Novel load prediction influence factor correlation analysis algorithm

Abstract

The invention discloses a novel load prediction influence factor correlation analysis algorithm, which comprises the following steps: a frequency spectrum identification step: acquiring influence factor dynamic time sequence data, forming a factor frequency spectrum, and identifying a first peak frequency and a second peak frequency at the high-frequency side of the factor frequency spectrum to determine a frequency band bandwidth; the parameter inversion step is used for obtaining factor equivalent correlation stiffness and slow-varying foundation correlation strength based on constraint conditions of the core factors and the correlation factors; and a main mode locking step. Relates to the technical field of load prediction. According to the method, influence factor dynamic time sequence data are acquired through a frequency spectrum identification step, a factor frequency spectrum is generated, a first peak frequency and a second peak frequency at a high-frequency side are identified to determine a frequency band bandwidth, and a precise frequency boundary is provided for subsequent correlation characteristic analysis; and inversely solving factor equivalent correlation stiffness and slow-varying foundation correlation strength through a parameter inversion step based on constraint conditions of a core factor and a correlation factor, and defining an action mechanism of influence of a factor correlation characteristic on a load.

Description

Technical Field

[0001] This invention relates to the field of load forecasting technology, and in particular to a novel algorithm for analyzing the correlation of load forecasting influencing factors. Background Technology

[0002] Load forecasting is widely used in power system dispatching, energy resource optimization, industrial production planning, and other fields. Its forecasting accuracy directly affects the operating efficiency, cost control, and safety and stability of related systems. During the load forecasting process, the factors influencing load changes are complex and diverse, including fluctuations in historical load data, changes in environmental parameters, and adjustments in user behavior patterns. These factors have complex interrelationships, and some factors exhibit dynamic fluctuations, thus indirectly affecting the accuracy of the load forecasting results.

[0003] Existing technologies mostly employ single-dimensional factor analysis methods, failing to fully consider the dynamic characteristics and interrelationships of each influencing factor. They are susceptible to interference factors that lead to biased analysis results, cannot accurately distinguish between the basic correlation strength and instantaneous dynamic correlation increment of core influencing factors, and ignore the impact of dynamic fluctuation factors on the correlation relationship. This results in unreasonable selection of input parameters for load forecasting models, insufficient forecast accuracy, and difficulty in meeting the needs of practical application scenarios. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies in load forecasting, where the dynamic fluctuations of influencing factors and changes in the correlation characteristics between factors make it difficult to accurately analyze the correlation between core influencing factors and load. Therefore, this invention proposes a novel correlation analysis algorithm for load forecasting influencing factors.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution:

[0006] A novel correlation analysis algorithm for load forecasting influencing factors includes:

[0007] The spectrum identification step is used to acquire dynamic time-series data of influencing factors and form factor spectrum, and to identify the first peak frequency and the second peak frequency on the high-frequency side of the factor spectrum in order to determine the frequency bandwidth.

[0008] The parametric inversion step is used to obtain the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations based on the constraints of core factors and related factors.

[0009] The main mode locking step is used to generate a set of associated modes based on the factor equivalent association stiffness and the slow-varying foundation association strength, and to determine the main associated mode based on the factor association slope energy concentration. The center frequency of the locked frequency band is determined by the main associated mode and the slow-varying foundation association strength.

[0010] The correlation increment calculation step is used to extract the dynamic time series data of influencing factors from the center frequency and bandwidth of the locked band to obtain the dynamic components of the locked band factors, and convert the dynamic components of the locked band factors into load influence increments based on the relationship between the correlation increment and the load influence increment in the correlation geometry and data correlation analysis.

[0011] The modal comparison step is used to set the analysis time window based on the locked frequency band center frequency and locked frequency band bandwidth, compare the main associated modes in the current analysis time window with the main associated modes in the previous analysis time window, and obtain the main associated mode comparison results;

[0012] The decision generation step is used to issue factor verification instructions based on the main correlation modality comparison results, and to determine the factor retesting priority level based on the load impact increment.

[0013] Preferably, the process involves acquiring dynamic time-series data of influencing factors and forming a factor spectrum, identifying a first peak frequency and a second peak frequency on the high-frequency side of the factor spectrum to determine the bandwidth, including:

[0014] Dynamic monitoring signals of influencing factors are collected, and constrained double integral processing is performed on the dynamic monitoring signals to obtain dynamic time series data of influencing factors;

[0015] Short-time Fourier transform is performed on the dynamic time-series data of influencing factors to generate factor spectra;

[0016] The first peak frequency and the second peak frequency are identified on the high-frequency side of the factor spectrum, and the frequency band boundaries are respectively defined by the nearest local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side for each peak frequency.

[0017] The difference between the local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side is used as the bandwidth.

[0018] Preferably, based on the constraints of core factors and related factors, the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations are obtained, including:

[0019] Based on the geometric correlation between the changes in factor correlation and dynamic data, as well as the physical equivalence of influence intensity and correlation degree, the factor correlation stiffness is equivalent to the correlation stiffness of the core correlation factor.

[0020] Based on the core factors and related factors, the association system model is established, and the boundary conditions of the core factors at the direct and indirect association ends are established, forming a mapping from intrinsic relationships and frequencies to association wavenumbers.

[0021] Substituting the first peak frequency and the second peak frequency into the eigenvalue relation respectively, we can form a joint constraint on the factor equivalent correlation stiffness and the slow-varying foundation correlation strength.

[0022] By minimizing the residuals under joint constraints, the equivalent correlation stiffness of the factors and the correlation strength of the slowly varying foundation are obtained.

[0023] Preferably, a set of associated modes is generated based on the equivalent correlation stiffness of the factors and the correlation strength of the slowly varying foundation, and the main correlation mode is determined based on the energy concentration of the factor correlation slope. The center frequency of the locked frequency band is determined by the main correlation mode and the correlation strength of the slowly varying foundation, including:

[0024] Based on the mapping between frequency and associated wavenumber, the set of associated wavenumbers that satisfy the intrinsic relation is obtained, and the associated mode set is obtained by sorting them in descending order;

[0025] Calculate the concentration of factor correlation slope energy for each of the associated modes, and select the mode with the highest concentration as the main associated mode.

[0026] The center frequency of the locked band is calculated based on the main correlation mode, the slowly varying basic correlation strength, and the factor correlation density.

[0027] Preferably, the dynamic components of the locked band factors are obtained by bandpass extraction of the dynamic time-series data of influencing factors based on the center frequency and bandwidth of the locked band. Then, based on the relationship between the correlation increment and the load impact increment in the geometric correlation analysis, the dynamic components of the locked band factors are converted into load impact increments, including:

[0028] The left and right band edges are calculated using the locked band center frequency and the band bandwidth. The left band edge is equal to the locked band center frequency minus half of the band bandwidth, and the right band edge is equal to the locked band center frequency plus half of the band bandwidth.

[0029] Bandpass filtering is applied between the left and right band edges of the dynamic time series data of influencing factors to obtain the dynamic components of the locked band factors.

[0030] Under weak interference conditions, the dynamic components of the locked-band factors are extrapolated to the dynamic distribution of the full-factor association using the spatial association function of the main association mode, and the association increment is obtained, which is half of the integral of the square of the first derivative of the dynamic distribution along the association dimension within the effective association range.

[0031] Based on the factor correlation elasticity coefficient, correlation influence area, and effective correlation range, the correlation increment is converted into load influence increment, thus obtaining the load influence increment time series.

[0032] Preferably, an analysis time window is set according to the center frequency and bandwidth of the locked frequency band, and the main correlated modes in the current analysis time window are compared with the main correlated modes in the previous analysis time window to obtain the main correlated mode comparison results, including:

[0033] The analysis time window length is taken as the reciprocal of the locked frequency band bandwidth, and the analysis time window step pitch is taken as half of the reciprocal of the locked frequency band center frequency, generating an analysis time window sequence arranged in chronological order;

[0034] Select the current analysis time window and the previous analysis time window from the analysis time window sequence, read the main correlation modes in the two windows respectively, and calculate the difference between the main correlation modes;

[0035] When the difference between the primary correlation modes is not zero, the alignment result of the primary correlation modes is determined to have changed; when the difference between the primary correlation modes is zero, the alignment result of the primary correlation modes is determined to have not changed.

[0036] Based on the main correlation mode comparison results, factor verification instructions are issued, and the factor retesting priority level is determined based on the load impact increment, including:

[0037] When the main correlation modality comparison result shows a change, a factor verification instruction is issued. The verification instruction includes: suspending the current load forecasting model training or forecasting execution process, verifying the data of influencing factors and checking the correlation relationship. When the main correlation modality comparison result shows no change, the verification instruction is to continue monitoring and data recording.

[0038] When the main correlation mode comparison result changes, the root mean square value of the load influence increment time series is used as the influence intensity measure, and the factor retesting priority level is determined based on the ratio of the influence intensity measure to the slow-varying basic correlation intensity.

[0039] Preferably, an analytical evidence package is formed based on the comparison results of factor equivalent correlation stiffness, slow-varying foundation correlation strength, main correlation mode, locked frequency band center frequency, load influence increment, and main correlation mode.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. This invention acquires dynamic time-series data of influencing factors and generates factor spectra through a spectrum identification step, identifies the first and second peak frequencies on the high-frequency side to determine the bandwidth, and provides accurate frequency boundaries for subsequent correlation characteristic analysis; then, through a parametric inversion step, based on the constraints of core factors and related factors, it inversely calculates the equivalent correlation stiffness and slowly varying foundation correlation strength of the factors, clarifying the mechanism by which the correlation characteristics of factors affect the load; the correlation increment calculation step extracts the dynamic components of the locked band factors in combination with the locked band parameters, and converts them into load impact increments based on the geometry of correlation relationships and the principle of data correlation analysis, thus realizing accurate analysis of the correlation relationships of load influencing factors.

[0042] 2. This invention determines the main correlation mode and obtains the center frequency of the locked frequency band by determining the slope energy concentration of the factor correlation, thus clarifying the dominant correlation mode of the load forecasting influencing factors. Based on the locked frequency band parameters, an analysis time window is set, and the main correlation modes within the continuous time window are compared to obtain the main correlation mode comparison results. When the correlation characteristics of the influencing factors change due to data anomalies, sudden changes in correlation relationships, etc., the main correlation mode will shift accordingly and be reflected as a non-zero main correlation mode difference. This allows for timely identification of problems such as factor correlation failure and sudden changes in the influencing mechanism.

[0043] 3. This invention dynamically issues factor verification instructions based on the main correlation modal comparison results and determines the factor retest priority level by combining the slow-varying basic correlation strength. It can prioritize the allocation of retest resources to factors with high priority levels, which not only realizes timely response to abnormal load forecasting influencing factors, but also optimizes the retest resource scheduling, providing a strong guarantee for improving load forecasting accuracy. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0045] Figure 1 This is a functional block diagram of a novel load forecasting influencing factor correlation analysis algorithm provided in an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0047] Example: This example provides a novel correlation analysis algorithm for load forecasting influencing factors, specifically including:

[0048] The spectrum identification step is used to acquire dynamic time-series data of influencing factors and form factor spectrum, and to identify the first peak frequency and the second peak frequency on the high-frequency side of the factor spectrum in order to determine the frequency bandwidth.

[0049] In an embodiment of the present invention, acquiring dynamic time-series data of influencing factors and forming a factor spectrum, identifying a first peak frequency and a second peak frequency on the high-frequency side of the factor spectrum to determine the bandwidth, includes:

[0050] Dynamic monitoring signals of influencing factors are collected, and constrained double integral processing is performed on the dynamic monitoring signals to obtain dynamic time series data of influencing factors;

[0051] Specifically, for various influencing factors in load forecasting, such as ambient temperature, humidity, historical load data, and user electricity consumption behavior data, corresponding dynamic monitoring equipment is deployed. The sampling frequency of the monitoring equipment is set to be no less than twice the highest frequency of the preset high-frequency peak to ensure complete acquisition of dynamic change information of influencing factors. The acquired dynamic monitoring signals are preprocessed to remove DC components and high-frequency noise. Preprocessing is achieved using a low-pass filter with a cutoff frequency 1.5 times the highest frequency of the preset high-frequency peak. The preprocessed dynamic monitoring signals undergo a first integration operation. An initial constraint is introduced during the integration process, namely, the rate of change of influencing factors at the initial moment is zero. At the same time, the integration result is corrected in real time through sliding window mean filtering to avoid cumulative errors. The obtained signal undergoes a second integration operation. The second integration also introduces an initial constraint, namely, the baseline value of influencing factors at the initial moment is zero. During the integration process, periodic calibration is performed by comparing with historical statistical baseline values ​​to ensure that the final dynamic time-series data of influencing factors can accurately reflect the dynamic change process of influencing factors.

[0052] Short-time Fourier transform is performed on the dynamic time-series data of influencing factors to generate factor spectra;

[0053] The first peak frequency and the second peak frequency are identified on the high-frequency side of the factor spectrum, and the frequency band boundaries are respectively defined by the nearest local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side for each peak frequency.

[0054] The difference between the local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side is taken as the bandwidth.

[0055] Specifically, the factor spectrum is a graph reflecting the energy distribution of dynamic time-series data of influencing factors at different frequencies over time; the first peak frequency and the second peak frequency are the two frequency components with the highest energy concentration and stability in the high-frequency side of the factor spectrum, and are key characteristic quantities reflecting the dynamic correlation characteristics of influencing factors; the bandwidth is a parameter describing the frequency distribution range of the first peak frequency and the second peak frequency in the factor spectrum, and is used to quantify the width of the frequency interval where the peak frequency is located.

[0056] Specifically, a rectangular window with a length 100 times the sampling period of the dynamic time-series data of influencing factors is selected as the analysis window for the short-time Fourier transform. The overlap rate of the window function is set to 50% to ensure a balance between time resolution and frequency resolution. The dynamic time-series data of influencing factors is segmented according to the analysis window length and overlap rate. A Fourier transform is performed on each segment to obtain the frequency components and amplitudes corresponding to each segment. The Fourier transform results of each segment are arranged in chronological order to form a factor spectrum with time as the horizontal axis, frequency as the vertical axis, and amplitude as grayscale or color intensity. The number of points in the Fourier transform is set to twice the length of the analysis window to improve frequency resolution, and the frequency range covers from 0 to half of the sampling frequency, ensuring that all frequency components contained in the dynamic time-series data of influencing factors and their time-varying characteristics can be fully presented.

[0057] Specifically, the region in the spectrum with frequencies higher than one-third of the sampling frequency is selected as the high-frequency side. Within this high-frequency side, the amplitude of each frequency point is scanned point by point along the frequency axis. When the amplitude of a frequency point is simultaneously greater than the amplitudes of its left and right adjacent frequency points, that frequency point is determined to be a local peak. For all identified local peaks, the amplitude fluctuation coefficient of each peak point on the entire time axis is calculated, where the fluctuation coefficient is the ratio of the standard deviation to the mean of the peak amplitude sequence. The two local peaks with the smallest fluctuation coefficients are selected as the first peak frequency and the second peak frequency, respectively. For the first peak frequency, Starting from the frequency position, check the amplitude of adjacent frequency points sequentially towards the lower frequency direction. When the amplitude of a certain frequency point is simultaneously less than the amplitude of the adjacent frequency point to its left and the amplitude of the adjacent frequency point to its right, determine that frequency point as the local minimum frequency on the low-frequency side of the first peak frequency. Then, starting from the frequency position of the first peak frequency, check the amplitude of adjacent frequency points sequentially towards the higher frequency direction, and similarly determine the local minimum frequency on the high-frequency side of the first peak frequency. In the same way, determine the local minimum frequency on the low-frequency side and the local minimum frequency on the high-frequency side of the second peak frequency respectively. Each local minimum frequency is the frequency band boundary of the corresponding peak frequency.

[0058] Specifically, for the first peak frequency, the identified local minimum frequencies on the high-frequency side and low-frequency side are extracted, and the local minimum frequencies on the low-frequency side are subtracted from the local minimum frequencies on the high-frequency side. The result is used as the bandwidth corresponding to the first peak frequency. For the second peak frequency, the bandwidth corresponding to the second peak frequency is obtained using the same method.

[0059] The parametric inversion step is used to obtain the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations based on the constraints of core factors and related factors.

[0060] In embodiments of the present invention, based on the constraints of core factors and related factors, the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations are obtained, including:

[0061] Based on the geometric correlation between the changes in factor correlation and dynamic data, as well as the physical equivalence of influence intensity and correlation degree, the factor correlation stiffness is equivalent to the correlation stiffness of the core correlation factor.

[0062] Based on the core factors and related factors, the association system model is established, and the boundary conditions of the core factors at the direct and indirect association ends are established, forming a mapping from intrinsic relationships and frequencies to association wavenumbers.

[0063] Substituting the first peak frequency and the second peak frequency into the eigenvalue relation respectively, we can form a joint constraint on the factor equivalent correlation stiffness and the slow-varying foundation correlation strength.

[0064] By minimizing the residual under joint constraints, the factor equivalent correlation stiffness and the correlation strength of the slowly varying foundation are obtained;

[0065] Specifically, factor correlation stiffness refers to the ability of the correlation between influencing factors to resist change. Its magnitude is equal to the influence intensity required to produce a unit change in the correlation, reflecting the stability of the correlation between factors. The correlation stiffness of the core correlation factor is a parameter used to equivalently simulate the factor correlation stiffness. Its physical meaning is the influence intensity required to produce a unit change in the correlation of the core correlation factor. The core factor specifically refers to the key influencing factor that plays a decisive role in the load forecasting results, such as historical load data and electricity consumption behavior data of major user groups. The correlation factor refers to the influencing factor that has a correlation with the core factor and indirectly affects the load forecasting results. The equivalent correlation stiffness of the factor is obtained by inverse calculation through dynamic monitoring signals and is a parameter that equivalently reflects the actual correlation stiffness between factors. The slow-varying basic correlation strength refers to the basic correlation strength between the core factor and the correlation factor, which changes slowly over time, and is different from the correlation increment caused by dynamic fluctuations. The correlation system model is a simplified model used to simulate the correlation characteristics between core factors and related factors in load forecasting. Its core is to abstract the factor correlation relationships in the actual scenario into a combined system of core factors and related factors. In this model, the core factor has definite parameters such as effective correlation range and correlation density. Its correlation behavior is affected by the basic correlation strength and its own characteristic distribution. The connection between the core factor and the related factor is equivalent to the correlation stiffness of the core related factor, which is used to simulate the constraint effect of the ability of the correlation between factors to resist changes on the overall correlation characteristics. The indirect correlation end of the core factor is usually set as a fixed constraint, that is, the correlation change at this position is zero. By establishing the influence balance condition of the direct correlation end and the correlation constraint condition of the indirect correlation end, the intrinsic relationship describing the inherent correlation characteristics of the system can be derived. This relationship reflects the inherent correlation between parameters such as correlation wavenumber, effective correlation range, basic correlation strength, and factor equivalent correlation stiffness.

[0066] Specifically, the geometric relationships of factor correlation changes are analyzed. When the correlation between factors changes, the relationship between the dynamic data change and the correlation change is determined by the factor correlation eccentricity. The dynamic data change is equal to the product of the correlation change and the factor correlation eccentricity. Simultaneously, the physical equilibrium relationship of the forces acting on the factor correlation is analyzed. The influence strength of the factors resisting the correlation change is equal to the product of the factor correlation stiffness and the correlation change. The influence strength of the core correlation factor hindering the correlation change is equal to the product of the core correlation factor's correlation stiffness and the dynamic data change. The influence moment generated by this influence strength on the correlation center is equal to the product of the influence strength and the factor correlation eccentricity. Based on the principle of influence moment balance, the influence moment of the factors resisting the correlation change is equal to the influence moment generated by the influence strength of the core correlation factor. Therefore, the equivalent relationship between the factor correlation stiffness and the core correlation factor's correlation stiffness is obtained: the correlation stiffness of the core correlation factor is equal to the factor correlation stiffness divided by the square of the factor correlation eccentricity.

[0067] Specifically, the core factor is considered as the associated subject, and its effective association range is defined as the association dimension length from the direct association end to the indirect association end. The basic association strength is the slowly varying basic association strength, and the association density is the association strength distribution per unit association dimension length. The core association factor is set at the direct association end, and its association stiffness is determined by the ratio of the factor association stiffness to the square of the association eccentricity. The direct association end is defined as the origin of the coordinate system, and the indirect association end is defined as the coordinate endpoint along the association dimension direction. An influence balance condition is established for the direct association end: the product of the association change slope of the core factor at the direct association end and the basic association strength is equal in magnitude and opposite in direction to the product of the association stiffness of the core association factor and the association change at the direct association end. A constraint condition is established for the indirect association end: the association change of the core factor at the indirect association end is zero. The association mode function of the core factor is set as a sine function, and its independent variable is the product of the association wavenumber and the distance from the indirect association end to the current association position. Substituting the mode function into the conditions of the direct and indirect association ends, the intrinsic relation is derived, where the formula for the intrinsic relation is:

[0068]

[0069] In the formula, k is the correlation wavenumber, used to characterize the spatial distribution characteristics of the correlation changes of the core factor; L is the effective correlation range of the core factor, referring to the correlation dimension length from the direct correlation end to the indirect correlation end; T is the slowly varying basic correlation strength, which is the basic correlation strength between the core factor and the related factors; F is the factor equivalent correlation stiffness, reflecting the ability of the correlation between factors to resist changes, and is a quantitative parameter of the factor correlation characteristics; e is the factor correlation eccentricity. This formula is derived based on the principle of factor correlation torque balance. The left side reflects the spatial distribution characteristics of the correlation changes of the core factor, and the right side reflects the influence of factor correlation constraints and basic correlation strength on the correlation changes. The balance between the two determines the correlation wavenumber value of the inherent correlation mode of the system.

[0070] Based on the correlation fluctuation characteristics of core factors, the core manifestation of these characteristics is that the propagation speed of the correlation fluctuation is jointly determined by the basic correlation strength and correlation density of the core factors. Furthermore, the fluctuation exhibits both temporal periodicity (manifested as frequency) and spatial periodicity (manifested as correlation wavelength), which are linked through the correlation wave velocity to form a unified fluctuation pattern. Specifically, under weak interference conditions, the correlation fluctuation of the core factors satisfies the wave equation, and the correlation wave velocity depends only on the basic correlation strength and correlation density, and is independent of the frequency and amplitude of the correlation fluctuation. Here, the correlation wave velocity v is equal to the square root of the ratio of the basic correlation strength T to the correlation density μ, i.e. The correlated wavenumber is a parameter describing the spatial periodicity of correlated fluctuations, and its relationship with the correlated wavelength is as follows: Where k is the associated wavenumber, λ is the associated wavelength, and the associated wavelength is the associated dimensional distance traveled by the associated wave within one period. The associated wave velocity v, frequency f (the number of associated wave periods per unit time), and associated wavelength λ satisfy the following relationship: From this, the associated wavelength can be derived. The associated wave speed expression Substituting into the above equation establishes the mapping from frequency f to the associated wavenumber k, i.e. This mapping reflects the quantitative correlation between the temporal periodicity parameter (frequency) and the spatial periodicity parameter (correlated wavenumber) of the associated fluctuations, providing a foundation for subsequent solutions to the associated wavenumber and associated modal characteristics through frequency characteristics.

[0071] Specifically, the frequency values ​​of the first peak frequency and the second peak frequency are determined, where the first peak frequency is the frequency corresponding to the local peak point with the smallest high-frequency side fluctuation coefficient, and the second peak frequency is the frequency corresponding to the local peak point with the second smallest high-frequency side fluctuation coefficient. The first peak frequency is substituted into the frequency-to-correlated wavenumber mapping relationship to obtain the correlated wavenumber corresponding to the first peak frequency. This correlated wavenumber is then substituted into the eigenvalue relation, where the tangent of the product of the correlated wavenumber and the effective correlation range is equal to the ratio of the product of the base correlation strength, the correlated wavenumber, and the square of the correlation eccentricity to the factor equivalent correlation stiffness. This yields the first peak frequency, which includes the factor equivalent correlation stiffness and the slowly varying base correlation strength. One equation; following the same method, substitute the second peak frequency into the frequency-to-correlated wavenumber mapping relationship to obtain the corresponding correlated wavenumber, and then substitute this correlated wavenumber into the eigenvalue relation to obtain the second equation containing the factor equivalent correlation stiffness and the slow-varying foundation correlation strength; the two equations together constitute the joint constraint on the factor equivalent correlation stiffness and the slow-varying foundation correlation strength, where the effective correlation range is the correlation dimension length of the core factor from the direct correlation end to the indirect correlation end, the correlation eccentricity is the distance from the factor correlation stress point to the correlation center, and the correlation density is the correlation strength distribution per unit correlation dimension length of the core factor.

[0072] Specifically, the initial search range for the factor equivalent correlation stiffness is set to zero to the preset maximum correlation stiffness, and the initial search range for the slowly varying foundation correlation strength is set to zero to the preset maximum foundation correlation strength. The preset maximum correlation stiffness and preset maximum foundation correlation strength are determined based on the characteristics of the load prediction influencing factors and the actual application scenario. Any set of factor equivalent correlation stiffness and slowly varying foundation correlation strength within the initial search range is selected as the initial values ​​for iteration. These are substituted into two equations constructed from the first and second peak frequencies, respectively. The difference between the left and right sides of each equation is calculated as the residual, and the sum of the squares of the two residuals is used as the objective function value. The gradient descent method is used to iteratively update the factor equivalent correlation stiffness and slowly varying foundation correlation strength. During each iteration, the parameter values ​​are adjusted according to the gradient direction of the objective function to reduce the objective function value. This iterative process is repeated until the objective function value is less than the preset residual threshold. At this point, the corresponding factor equivalent correlation stiffness and slowly varying foundation correlation strength are the optimal solutions under joint constraints. The preset residual threshold is set according to the analysis accuracy requirements to ensure that the solution results meet the accuracy requirements of subsequent analyses.

[0073] The main mode locking step is used to generate a set of associated modes based on the factor equivalent association stiffness and the slow-varying foundation association strength, and to determine the main associated mode based on the factor association slope energy concentration. The center frequency of the locked frequency band is determined by the main associated mode and the slow-varying foundation association strength.

[0074] In embodiments of the present invention, a set of associated modes is generated based on the factor equivalent association stiffness and the slowly varying foundation association strength, and the main associated mode is determined based on the factor association slope energy concentration. The center frequency of the locked frequency band is determined by the main associated mode and the slowly varying foundation association strength, including:

[0075] Based on the mapping between frequency and associated wavenumber, the set of associated wavenumbers that satisfy the intrinsic relation is obtained, and the associated mode set is obtained by sorting them in descending order;

[0076] Calculate the concentration of factor correlation slope energy for each of the associated modes, and select the mode with the highest concentration as the main associated mode.

[0077] The center frequency of the locked frequency band is calculated based on the main correlation mode, the slowly varying basic correlation strength, and the factor correlation density.

[0078] Specifically, the associated mode set is a collection of all associated wavenumbers that satisfy intrinsic relationships in the system of core factors and associated factors. Each associated wavenumber corresponds to an intrinsic associated mode of the system, reflecting the fluctuation characteristics of the system under different associated modes. The factor associated slope energy concentration is a parameter used to quantify the degree of energy concentration of a certain associated mode on the associated slope at the direct associated end. The larger the value, the more concentrated the energy of the associated mode is on the associated slope change at the direct associated end. The main associated mode is the associated mode with the largest factor associated slope energy concentration in the associated mode set, and can most significantly reflect the associated characteristics between the core factors and associated factors. The locked frequency band center frequency is the center frequency of the bandpass filter used to determine the subsequent extraction of the dynamic components of factors, and is used to lock the associated fluctuation frequency components corresponding to the main associated mode.

[0079] Specifically, based on the obtained factor equivalent correlation stiffness and slow-varying foundation correlation strength, the search range of correlation wavenumbers is set from zero to a preset maximum correlation wavenumber. The preset maximum correlation wavenumber is determined based on the effective correlation range of the core factor and the highest analysis frequency. Within the search range, the Newton-Raphson iteration method is used to solve the intrinsic relation equation, i.e., the tangent of the product of correlation wavenumber and effective correlation range is equal to the ratio of the product of slow-varying foundation correlation strength, correlation wavenumber, and the square of correlation eccentricity to factor equivalent correlation stiffness. All positive roots satisfying this equation are obtained, forming a set of correlation wavenumbers. All correlation wavenumbers in the set are arranged in descending order of value, and the sorted set of correlation wavenumbers is the set of correlation modes. Each correlation wavenumber corresponds to an intrinsic correlation mode of the correlation system between the core factor and the related factors.

[0080] Specifically, iterate through each association mode in the association mode set. For each association mode's corresponding wavenumber, determine its association mode function as a sine function. The independent variable of this function is the product of the association wavenumber and the distance from a certain association position on the core factor to the indirect association end. Calculate the slope of the association mode function at the direct association end, i.e., take the first derivative of the association mode function at the direct association end position (the origin) to obtain the association slope value at the direct association end. Square the association slope value at the direct association end to obtain the association slope energy at the direct association end. Calculate the association... The total energy of a mode is half the product of the square of the correlation mode function over the entire effective correlation range of the core factor and the correlation density and correlation velocity, where the correlation velocity is the square root of the ratio of the slowly varying basic correlation strength to the correlation density. The correlation slope energy at the direct correlation end is divided by the total energy of the correlation mode to obtain the concentration of factor correlation slope energy at the direct correlation end of the correlation mode. The concentration of factor correlation slope energy of all correlation modes in the correlation mode set is compared, and the correlation mode with the largest concentration value is selected as the main correlation mode.

[0081] Specifically, the associated wavenumbers corresponding to the main associated modes are extracted, and the associated wave velocity is calculated based on the slowly varying basic association strength and association density. The associated wave velocity is the square root of the ratio of the slowly varying basic association strength to the association density, and the association density is the association strength distribution per unit association dimension of the core factor. The angular frequency is calculated based on the product of the associated wavenumbers and the associated wave velocity, and then the angular frequency is divided by twice pi. The result is the center frequency of the locked frequency band.

[0082] The correlation increment calculation step is used to extract the dynamic time series data of influencing factors from the center frequency and bandwidth of the locked band to obtain the dynamic components of the locked band factors, and convert the dynamic components of the locked band factors into load influence increments based on the relationship between the correlation increment and the load influence increment in the correlation geometry and data correlation analysis.

[0083] In an embodiment of the present invention, bandpass extraction is performed on the dynamic time-series data of influencing factors based on the center frequency and bandwidth of the locked band to obtain the dynamic components of the locked band factors. Then, based on the relationship between the correlation increment and the load impact increment in the correlation geometry and data correlation analysis, the dynamic components of the locked band factors are converted into load impact increments, including:

[0084] The left and right band edges are calculated using the locked band center frequency and the band bandwidth. The left band edge is equal to the locked band center frequency minus half of the band bandwidth, and the right band edge is equal to the locked band center frequency plus half of the band bandwidth.

[0085] Bandpass filtering is applied between the left and right band edges of the dynamic time series data of influencing factors to obtain the dynamic components of the locked band factors.

[0086] Under weak interference conditions, the dynamic components of the locked-band factors are extrapolated to the dynamic distribution of the full-factor association using the spatial association function of the main association mode, and the association increment is obtained, which is half of the integral of the square of the first derivative of the dynamic distribution along the association dimension within the effective association range.

[0087] Based on the factor correlation elasticity coefficient, correlation influence area and effective correlation range, the correlation increment is converted into load influence increment, and the load influence increment time series is obtained.

[0088] Specifically, the weak interference condition refers to the small dynamic fluctuation of the influencing factors. At this time, the correlation change between the core factor and the related factors is small, and the influence of the nonlinearity of the correlation change on the geometry of the correlation relationship can be ignored. The deformation state of the correlation can be described by linearization alone. The correlation increment is the change in the correlation length of the core factor due to the dynamic fluctuation, reflecting the incremental change in the correlation dimension length of the core factor during the correlation fluctuation process. The load impact increment is the change in the load due to the correlation increment. Arranging the load impact increments at different times in chronological order forms the load impact increment time series, which reflects the dynamic change of the load impact over time.

[0089] Specifically, the calculated center frequency of the locked band and the bandwidth corresponding to the peak frequency associated with the main associated mode are extracted. Half of the bandwidth is calculated, i.e., the bandwidth is divided by two. The center frequency of the locked band is subtracted from half of the bandwidth, and the result is taken as the left band edge. The center frequency of the locked band is added to half of the bandwidth, and the result is taken as the right band edge. The left and right band edges together define the frequency range of the locked band. A Butterworth bandpass filter is selected as the filtering tool. The passband range of this filter is defined by the left and right band edges. The filter order is set to fourth order to balance filtering accuracy and signal distortion. Based on the sampling frequency of the left and right band edges and the dynamic time-series data of influencing factors, the cutoff frequency parameter of the filter is calculated. The cutoff frequency parameter is calculated by dividing the left and right band edges. The ratio of the sampling frequency to half is used; the dynamic time series data of influencing factors are preprocessed to remove DC components and obvious noise interference, ensuring the continuity and integrity of the time series data. The preprocessed dynamic time series data of influencing factors is input into the designed Butterworth bandpass filter. First, forward filtering is performed, and then reverse filtering is performed on the filtering result to eliminate the phase shift generated during the filtering process. The signal after bidirectional filtering is extracted, which is the dynamic component of the locked band factor. This component only contains the frequency components between the left and right band edges. The spectrum analysis of the dynamic component of the locked band factor is performed to verify whether its frequency components are completely within the range defined by the left and right band edges. If there are frequency components outside the range, the filter order is adjusted and filtering is performed again until the frequency range requirements are met.

[0090] Specifically, the spatial correlation function of the main correlation mode is determined to be a sine function, with the independent variable being the product of the correlation wavenumber corresponding to the main correlation mode and the distance from a certain correlation position on the core factor to the indirect correlation end. The dynamic component of the locked-band factor is extracted; this component is the signal containing only the frequency components between the left and right band edges after bandpass filtering of the dynamic time-series data of the influencing factors. Under weak interference conditions, the dynamic distribution of the full-factor correlation is expressed as the product of the ratio of the dynamic component of the locked-band factor to the spatial correlation function at that correlation position and the spatial correlation function itself. The ratio of the spatial correlation function at that correlation position represents the ratio of the spatial correlation function at any point in the full-factor correlation. The ratio of the function value at the position of the correlation to the function value at the direct correlation end of the spatial correlation function is used to extrapolate from the dynamic components of the factor to the dynamic distribution of the correlation of all factors. The first derivative of the dynamic distribution of the correlation of all factors with respect to the correlation dimension is obtained to obtain the rate of change of the correlation along the correlation dimension. The rate of change is squared and then integrated over the effective correlation range of the core factor (the length of the correlation dimension from the direct correlation end to the indirect correlation end). The integration process adopts the numerical integration method. The result of the integration is multiplied by one half, and the value obtained is the correlation increment. This increment reflects the change in the length of the correlation dimension caused by the dynamic fluctuation of the core factor.

[0091] Specifically, the factor correlation elasticity coefficient and correlation influence area are extracted. The factor correlation elasticity coefficient is an inherent attribute of the relationship between the core factor and related factors, determined by the factor correlation characteristics. The correlation influence area is the range of the influence of the correlation on the load, determined by the actual application scenario. The effective correlation range of the core factor is extracted, which is the correlation dimension length from the direct correlation end to the indirect correlation end. For each time value in the correlation increment time series, the correlation strain at that time is calculated. The correlation strain is equal to the correlation increment at that time divided by the effective correlation range of the core factor. According to the correlation influence principle, the load influence increment is equal to the product of the factor correlation elasticity coefficient, the correlation influence area, and the correlation strain. That is, for the correlation strain at each time, it is multiplied by the factor correlation elasticity coefficient and the correlation influence area to obtain the load influence increment at that time. The load influence increments at all times are arranged in chronological order to form the load influence increment time series.

[0092] The modal comparison step is used to set the analysis time window based on the locked frequency band center frequency and locked frequency band bandwidth, compare the main associated modes in the current analysis time window with the main associated modes in the previous analysis time window, and obtain the main associated mode comparison results;

[0093] In embodiments of the present invention, an analysis time window is set according to the locked frequency band center frequency and the locked frequency band bandwidth. The main associated mode in the current analysis time window is compared with the main associated mode in the previous analysis time window to obtain the main associated mode comparison result, including:

[0094] The analysis time window length is taken as the reciprocal of the locked frequency band bandwidth, and the analysis time window step pitch is taken as half of the reciprocal of the locked frequency band center frequency, generating an analysis time window sequence arranged in chronological order;

[0095] Select the current analysis time window and the previous analysis time window from the analysis time window sequence, read the main correlation modes in the two windows respectively, and calculate the difference between the main correlation modes;

[0096] When the difference between the main associated modes is not zero, the main associated mode comparison result is determined to have changed; when the difference between the main associated modes is zero, the main associated mode comparison result is determined to have not changed.

[0097] Specifically, the determined locked frequency band bandwidth is extracted, and the length of the analysis time window is calculated, which is equal to the reciprocal of the locked frequency band bandwidth; the center frequency of the locked frequency band is extracted, and the step pitch of the analysis time window is calculated, which is equal to half the reciprocal of the center frequency of the locked frequency band; the total duration of the dynamic time series data of the influencing factors is determined, which is the product of the number of sampling points and the sampling interval of the time series data. The zero point of time is taken as the start time of the first analysis time window, and the start time of each analysis time window is the sum of the start time of the previous analysis time window and the step pitch, and the end time is the sum of the corresponding start time and the length of the analysis time window; the analysis time windows are repeatedly generated until the end time of a certain analysis time window is greater than or equal to the total duration of the dynamic time series data of the influencing factors, at which point the generation stops; all generated analysis time windows are arranged in ascending order of their start times to form a sequence of analysis time windows arranged in chronological order.

[0098] Specifically, the latest time window in the chronologically ordered analysis time window sequence is selected as the current analysis time window, and the previous one is selected as the previous analysis time window. The time interval between the two windows is the analysis time window step pitch. For the current analysis time window and the previous analysis time window, the correlation wavenumbers corresponding to their main correlation modes are extracted. The correlation wavenumbers of the current main correlation mode are subtracted from the correlation wavenumbers of the previous main correlation mode to obtain the main correlation mode difference. If the main correlation mode difference is not equal to zero, the main correlation mode comparison result is determined to have changed; if the main correlation mode difference is equal to zero, the main correlation mode comparison result is determined to have not changed.

[0099] The decision generation step is used to issue factor verification instructions based on the main correlation modality comparison results, and to determine the factor retesting priority level based on the load impact increment;

[0100] In an embodiment of the present invention, a factor verification instruction is issued based on the main correlation mode comparison results, and the factor retest priority level is determined based on the load impact increment, including:

[0101] When the main correlation modality comparison result shows a change, a factor verification instruction is issued. The verification instruction includes: suspending the current load forecasting model training or forecasting execution process, verifying the data of influencing factors and checking the correlation relationship. When the main correlation modality comparison result shows no change, the verification instruction is to continue monitoring and data recording.

[0102] When the main correlation mode comparison result changes, the root mean square value of the load influence increment time series is used as the influence intensity measure, and the factor retesting priority level is determined based on the ratio of the influence intensity measure to the slow-varying basic correlation intensity.

[0103] Based on the comparison results of factor equivalent correlation stiffness, slow-varying foundation correlation strength, main correlation mode, locked frequency band center frequency, load influence increment, and main correlation mode, an analysis evidence package is formed;

[0104] Specifically, if the main correlation modality comparison result shows a change, a factor verification instruction is immediately generated and issued. This instruction includes three operations: First, pausing the currently ongoing load forecasting model training or forecasting execution process, stopping all load forecasting-related operations; second, verifying the influencing factor data, using data consistency verification methods to verify the dynamic monitoring data of the influencing factors to ensure data accuracy and completeness; and third, verifying the correlation relationship, using data analysis tools to check whether the correlation between core factors and related factors has changed abnormally, and recording any problems found during the verification process. If the main correlation modality comparison result shows no change, a verification instruction is generated and issued. This instruction requires continued monitoring and data recording, i.e., continuously collecting dynamic time-series data of influencing factors, calculating and storing the main correlation modality parameters for each analysis time window in real time, and maintaining dynamic monitoring of the main correlation modality comparison results until the comparison results for the next analysis time window are generated.

[0105] Specifically, when the determination result is a change, the load impact increment time series corresponding to the changed analysis time window is extracted. The load impact increment value at each moment in the time series is squared, and the arithmetic mean of all squared values ​​within the time window is calculated. Then, the square root of this mean is taken, and the resulting value is the impact intensity metric. This impact intensity metric represents the overall intensity of the dynamic impact energy of the influencing factor on the load within the locked frequency band. The slowly varying basic correlation strength obtained by minimizing the joint constraint residual is extracted, and the ratio of the impact intensity metric to the slowly varying basic correlation strength is calculated to obtain a dimensionless ratio result. The larger the dimensionless value, the higher the retesting demand of the influencing factor under the current load forecasting scenario. The dimensionless ratio results of the influencing factors are sorted in descending order, and the verification instruction, the factor retesting priority level, time window number, and time index are written into the scheduling record unit for data backhaul and retesting scheduling. Resources and manual retesting resources are preferentially allocated to objects with higher factor retesting priority levels.

[0106] Specifically, to ensure full traceability, parameters such as factor equivalent correlation stiffness, slow-varying foundation correlation strength, main correlation mode, locked frequency band center frequency, load influence increment, main correlation mode comparison results, factor retest optimization level, and corresponding analysis time window are packaged into an analysis evidence package for use in subsequent before-and-after comparisons and quality acceptance after the retest is completed.

[0107] Specifically, to ensure full traceability, parameters such as node equivalent angular stiffness, slowly varying axial base tension, principal modes, locked frequency band center frequency, axial tensile force increment, principal mode comparison results, node retest optimization level, and corresponding analysis time window are packaged into an evidence package for subsequent comparison and quality acceptance after retesting.

[0108] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A novel correlation analysis algorithm for load forecasting influencing factors, characterized in that, include: The spectrum identification step is used to acquire dynamic time-series data of influencing factors and form factor spectrum, and to identify the first peak frequency and the second peak frequency on the high-frequency side of the factor spectrum in order to determine the frequency bandwidth. The parametric inversion step is used to obtain the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations based on the constraints of core factors and related factors. The main mode locking step is used to generate a set of associated modes based on the factor equivalent association stiffness and the slow-varying foundation association strength, and to determine the main associated mode based on the factor association slope energy concentration. The center frequency of the locked frequency band is determined by the main associated mode and the slow-varying foundation association strength. The correlation increment calculation step is used to extract the dynamic time series data of influencing factors from the center frequency and bandwidth of the locked band to obtain the dynamic components of the locked band factors, and convert the dynamic components of the locked band factors into load influence increments based on the relationship between the correlation increment and the load influence increment in the correlation geometry and data correlation analysis. The modal comparison step is used to set the analysis time window based on the locked frequency band center frequency and locked frequency band bandwidth, compare the main associated modes in the current analysis time window with the main associated modes in the previous analysis time window, and obtain the main associated mode comparison results; The decision generation step is used to issue factor verification instructions based on the main correlation modality comparison results, and to determine the factor retesting priority level based on the load impact increment.

2. The novel load forecasting influencing factor correlation analysis algorithm according to claim 1, characterized in that, Acquire dynamic time-series data of influencing factors and form a factor spectrum. Identify the first and second peak frequencies on the high-frequency side of the factor spectrum to determine the bandwidth, including: Dynamic monitoring signals of influencing factors are collected, and constrained double integral processing is performed on the dynamic monitoring signals to obtain dynamic time series data of influencing factors; Short-time Fourier transform is performed on the dynamic time-series data of influencing factors to generate factor spectra; The first peak frequency and the second peak frequency are identified on the high-frequency side of the factor spectrum, and the frequency band boundaries are respectively defined by the nearest local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side for each peak frequency. The difference between the local minimum frequency on the high-frequency side and the local minimum frequency on the low-frequency side is used as the bandwidth.

3. The novel load forecasting influencing factor correlation analysis algorithm according to claim 1, characterized in that, Based on the constraints of core factors and related factors, the equivalent correlation stiffness of factors and the correlation strength of slowly varying foundations are obtained, including: Based on the geometric correlation between factor correlation changes and dynamic data, as well as the physical equivalence of influence intensity and correlation degree, the factor correlation stiffness is equivalent to the correlation stiffness of the core correlation factor. Based on the core factors and related factors, the association system model is established, and the boundary conditions of the core factors at the direct and indirect association ends are established, forming a mapping from intrinsic relationships and frequencies to association wavenumbers. Substituting the first peak frequency and the second peak frequency into the eigenvalue relation respectively, we can form a joint constraint on the factor equivalent correlation stiffness and the slow-varying foundation correlation strength. By minimizing the residuals under joint constraints, the equivalent correlation stiffness of the factors and the correlation strength of the slowly varying foundation are obtained.

4. The novel load forecasting influencing factor correlation analysis algorithm according to claim 3, characterized in that, A set of associated modes is generated based on the equivalent correlation stiffness of the factors and the correlation strength of the slowly varying foundation. The main correlation mode is determined based on the energy concentration of the factor correlation slope. The center frequency of the locked frequency band is determined by the main correlation mode and the correlation strength of the slowly varying foundation, including: Based on the mapping between frequency and associated wavenumber, the set of associated wavenumbers that satisfy the intrinsic relation is obtained, and the associated mode set is obtained by sorting them in descending order; Calculate the concentration of factor correlation slope energy for each of the associated modes, and select the mode with the highest concentration as the main associated mode. The center frequency of the locked band is calculated based on the main correlation mode, the slowly varying basic correlation strength, and the factor correlation density.

5. The novel load forecasting influencing factor correlation analysis algorithm according to claim 1, characterized in that, Based on the center frequency and bandwidth of the locked band, bandpass extraction is performed on the dynamic time-series data of influencing factors to obtain the dynamic components of the locked band factors. Then, based on the relationship between the correlation increment and the load impact increment in the geometric correlation and data correlation analysis, the dynamic components of the locked band factors are converted into load impact increments, including: The left and right band edges are calculated using the locked band center frequency and the band bandwidth. The left band edge is equal to the locked band center frequency minus half of the band bandwidth, and the right band edge is equal to the locked band center frequency plus half of the band bandwidth. Bandpass filtering is applied between the left and right band edges of the dynamic time series data of influencing factors to obtain the dynamic components of the locked band factors. Under weak interference conditions, the dynamic components of the locked-band factors are extrapolated to the dynamic distribution of the full-factor association using the spatial association function of the main association mode, and the association increment is obtained, which is half of the integral of the square of the first derivative of the dynamic distribution along the association dimension within the effective association range. Based on the factor correlation elasticity coefficient, correlation influence area, and effective correlation range, the correlation increment is converted into load influence increment, thus obtaining the load influence increment time series.

6. The novel load forecasting influencing factor correlation analysis algorithm according to claim 1, characterized in that, The analysis time window is set according to the locked frequency band center frequency and locked frequency band bandwidth. The main correlated modes in the current analysis time window are compared with the main correlated modes in the previous analysis time window to obtain the main correlated mode comparison results, including: The analysis time window length is taken as the reciprocal of the locked frequency band bandwidth, and the analysis time window step pitch is taken as half of the reciprocal of the locked frequency band center frequency, generating an analysis time window sequence arranged in chronological order; Select the current analysis time window and the previous analysis time window from the analysis time window sequence, read the main correlation modes in the two windows respectively, and calculate the difference between the main correlation modes; When the difference between the main associated modes is not zero, the main associated mode comparison result is determined to have changed; when the difference between the main associated modes is zero, the main associated mode comparison result is determined to have not changed.

7. The novel load forecasting influencing factor correlation analysis algorithm according to claim 6, characterized in that, Based on the main correlation mode comparison results, factor verification instructions are issued, and the factor retesting priority level is determined based on the load impact increment, including: When the main correlation modality comparison result shows a change, a factor verification instruction is issued. The verification instruction includes: suspending the current load forecasting model training or forecasting execution process, verifying the data of influencing factors and checking the correlation relationship. When the main correlation modality comparison result shows no change, the verification instruction is to continue monitoring and data recording. When the main correlation mode comparison result changes, the root mean square value of the load influence increment time series is used as the influence intensity measure, and the factor retesting priority level is determined based on the ratio of the influence intensity measure to the slow-varying basic correlation intensity.

8. The novel load forecasting influencing factor correlation analysis algorithm according to claim 1, characterized in that, Based on the comparison results of factor equivalent correlation stiffness, slow-varying foundation correlation strength, main correlation mode, locked frequency band center frequency, load influence increment, and main correlation mode, an analysis evidence package is formed.

Images