A power equipment resource configuration optimization method based on intelligent analysis

Abstract

This invention relates to the field of power load forecasting technology, and in particular to a method for optimizing power equipment resource allocation based on intelligent analysis. This method converts historical load data sequences of a power system to the frequency domain to obtain a frequency richness assessment value. Based on the frequency richness assessment value, it obtains the adaptive component number of the VMD algorithm and decomposes the historical load data sequence into multiple component signals. Based on the load acquisition value and load forecast value of data points within the prediction analysis interval of any component signal, it performs prediction error analysis on any component signal at different preset quantiles to obtain an overall error value. Based on the energy intensity within the prediction analysis interval, it obtains the prediction influence weight of any component signal. It obtains the load forecast value corresponding to each component signal and combines the overall error value and prediction influence weight of each component signal to obtain a load forecast value interval, which is used for power equipment resource allocation optimization, significantly improving the accuracy and stability of load forecasting.

Description

Technical Field

[0001] This invention relates to the field of power load forecasting technology, and in particular to a method for optimizing power equipment resource allocation based on intelligent analysis. Background Technology

[0002] The power system, comprised of generation, transmission, transformation, distribution, and consumption, is a crucial foundation for achieving efficient power production, stable transmission, and secure supply. During power system operation, various power equipment serves as core resources, and their rational configuration directly impacts power supply reliability and operational efficiency. Current technologies use load forecasting models to predict load and allocate power equipment resources based on these predictions. However, load data typically exhibits multi-scale characteristics, including long-term trends, periodic fluctuations, and short-term random disturbances. The patterns of change differ significantly across time scales. Existing load forecasting models are usually based on load data at a single time scale, making it difficult to simultaneously depict both long-term evolution trends and short-term fluctuations. This results in load forecast data lacking a quantitative representation of uncertainty, failing to accurately reflect actual operational fluctuation risks. Consequently, when allocating power equipment resources based on load forecasts, it is difficult to balance operational safety and economy, and it cannot provide a reliable basis for redundant resource allocation and risk control, thus affecting the rationality of decision-making.

[0003] Therefore, there is an urgent need for a load forecasting method that can adaptively analyze multi-scale load characteristics and has the ability to assess forecast uncertainty, so as to improve the support level for resource allocation optimization and the stability of system operation. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a power equipment resource allocation optimization method based on intelligent analysis to solve the problem that the load forecasting results of the load forecasting model in the prior art lack quantitative expression of uncertainty and cannot truly reflect the fluctuation risk in actual operation.

[0005] This invention provides a method for optimizing power equipment resource allocation based on intelligent analysis, the method comprising the following steps: Load data is collected from the power system to obtain a historical load data sequence up to the current moment. The historical load data sequence is converted to the frequency domain to obtain a corresponding spectrum distribution map. The frequency richness of the spectrum distribution map is evaluated to obtain a frequency richness evaluation value. Based on the frequency richness evaluation value, the adaptive number of components is obtained when decomposing the historical load data sequence using the VMD algorithm, and the historical load data sequence is decomposed into multiple component signals. For any component signal, a prediction analysis interval is set for the any component signal. Based on the difference between the load acquisition value and the load prediction value of each data point within the prediction analysis interval, prediction error analysis is performed on the any component signal to obtain the error deviation degree under different preset quantiles. The overall error value is obtained by combining the error deviation degree under different preset quantiles. The prediction influence weight of the any component signal is obtained based on the energy intensity within the prediction analysis interval. The load forecast value corresponding to each component signal is obtained, and combined with the overall error value and prediction influence weight of each component signal, the load forecast value range after the current time is obtained, which is used for power equipment resource allocation optimization.

[0006] Preferably, the step of evaluating the frequency richness of the spectral distribution map to obtain a frequency richness evaluation value includes: The frequencies in the spectrum distribution map are linearly normalized to obtain normalized frequencies. The normalized frequencies are then combined in pairs. Based on the amplitude corresponding to each normalized frequency, the minimum amplitude in each combination is obtained. Based on the proportion of the minimum amplitude in each combination to the minimum amplitude in all combinations, the frequency contribution weight of each combination is obtained. The larger the proportion, the larger the corresponding frequency contribution weight. Calculate the absolute value of the difference between the two normalized frequencies contained in each combination, and perform a weighted average of all the absolute values ​​of the difference based on the frequency contribution weight of each combination to obtain the frequency richness evaluation value.

[0007] Preferably, obtaining the adaptive number of components when decomposing the historical load data sequence using the VMD algorithm based on the frequency richness evaluation value includes: Calculate the difference between the preset maximum number of components and the minimum number of components. Use the product of the difference and the frequency richness evaluation value as the quantity adjustment amount. Round down the sum of the minimum number of components and the quantity adjustment amount to obtain the adaptive number of components when decomposing the historical load data sequence using the VMD algorithm.

[0008] Preferably, the step of performing prediction error analysis on any component signal based on the difference between the load acquisition value and the load prediction value of each data point within the prediction analysis interval, to obtain the degree of error deviation under different preset quantiles, includes: Calculate the absolute value of the difference between the load collection value and the load prediction value for each data point within the prediction analysis interval, and record it as the prediction error for the corresponding data point, thereby obtaining the prediction error sequence; based on the maximum and minimum values ​​in the prediction error sequence, obtain the theoretical error value corresponding to each preset quantile; For any preset quantile, calculate the absolute value of the difference between the theoretical error value corresponding to the preset quantile and each prediction error in the prediction error sequence, and obtain the average absolute value of the difference, which is denoted as the error deviation degree under the preset quantile.

[0009] Preferably, the step of obtaining the overall error value by combining the error deviation degree under different preset quantiles includes: Based on the degree of error deviation at each preset quantile, a weight is assigned to the theoretical error value corresponding to each preset quantile. The greater the degree of error deviation, the smaller the weight. The theoretical error values ​​at all preset quantiles are then weighted and averaged to obtain the overall error value.

[0010] Preferably, obtaining the prediction influence weight of any component signal based on the energy intensity within the prediction analysis interval includes: Based on the load collection value of each data point within the prediction analysis interval, the accumulated load collection value is obtained and recorded as the energy intensity of the prediction analysis interval of any component signal. Based on the proportion of the energy intensity of the prediction analysis interval of any component signal in the energy intensity of all component signals' prediction analysis intervals, the prediction influence weight of any component signal is obtained. The larger the proportion, the greater the corresponding prediction influence weight.

[0011] Preferably, the step of obtaining the load forecast value corresponding to each component signal, and combining the overall error value and prediction influence weight of each component signal to obtain the load forecast value range after the current time includes: For any component signal, the any component signal is used as the input of the trained load forecasting model, and the corresponding output is the load forecast value after the current time. The sum of the load forecast value and the overall error value of the any component signal is calculated and recorded as the upward adjustment load forecast value of the any component signal. The subtraction between the load forecast value and the overall error value of the any component signal is calculated and recorded as the downward adjustment load forecast value of the any component signal. Based on the prediction influence weight of each component signal, the upper adjustment load prediction value of all component signals is weighted and averaged to obtain the upper limit of the load prediction after the current time; the lower adjustment load prediction value of all component signals is weighted and averaged to obtain the lower limit of the load prediction after the current time. The load forecast range after the current time is formed by the upper limit and lower limit of the load forecast after the current time.

[0012] Preferably, the prediction analysis interval of any component signal refers to the last m% of data points in any component signal, where m is a positive integer.

[0013] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: In this invention, a frequency richness assessment value is obtained by performing frequency domain transformation on historical load data sequences based on their spectral distribution characteristics. This assessment value is then used to adaptively adjust the number of components in variational mode decomposition (VMD), achieving multi-scale decomposition of low-frequency trends, mid-frequency cycles, and high-frequency disturbances. Then, for each component signal of the historical load data sequence, prediction error analysis is performed at different quantiles based on the difference between the load acquisition value and the predicted load value at each data point within the prediction analysis interval. This avoids the inadequacy of a single statistic in describing the prediction error characteristics, making the prediction error analysis more detailed. Through weighted processing, the overall error value of each component signal is obtained, ensuring that the error analysis more closely reflects the actual situation. This method effectively improves the representativeness and reliability of error assessment. Based on the energy intensity within the prediction analysis interval of each component signal, a prediction influence weight is set for each component signal. Combining the load prediction value of each component signal with the overall error value, a load prediction value interval for the current moment is constructed. This enables the characterization of the uncertainty of load changes, allowing power equipment resource allocation based on the load prediction value interval to balance safety margin and economy. While significantly improving the accuracy and stability of load forecasting, it provides a reliable basis for generator unit combination, transmission and distribution capacity planning, and redundancy decisions. This effectively reduces the risk of resource waste or insufficient supply caused by prediction deviations, and improves the overall reliability and economy of the power system operation. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for optimizing power equipment resource allocation based on intelligent analysis, provided in Embodiment 1 of the present invention. Detailed Implementation

[0016] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0017] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0018] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0019] See Figure 1 This is a flowchart of a power equipment resource allocation optimization method based on intelligent analysis provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include: Step S101: Load data of the power system is collected to obtain the historical load data sequence up to the current time. The historical load data sequence is converted to the frequency domain to obtain the corresponding spectrum distribution map. The frequency richness of the spectrum distribution map is evaluated to obtain the frequency richness evaluation value. Based on the frequency richness evaluation value, the number of adaptive components is obtained when decomposing the historical load data sequence using the VMD algorithm, and the historical load data sequence is decomposed into multiple component signals.

[0020] When optimizing the allocation of power equipment resources in a power system, the accuracy of power system load forecasting is crucial. Therefore, in this embodiment of the invention, a power load forecasting method that takes into account multiple time scales and has the ability to express uncertainty is constructed to improve the accuracy and reliability of power equipment resource allocation optimization decisions.

[0021] First, based on a preset sampling frequency, load data is collected from the power system, prioritizing historical load data from at least one year to obtain a historical load data sequence up to the current moment. This sequence fully reflects the daily, weekly, and seasonal variations of the power system. To ensure the continuity and reliability of the historical load data sequence, preprocessing operations such as cubic spline interpolation and Kalman filtering are performed to eliminate the influence of missing values ​​or outliers, providing a high-quality data foundation for subsequent frequency analysis, multi-scale decomposition, and predictive modeling. It is worth noting that the subsequent historical load data sequences are the preprocessed sequences.

[0022] In the process of load forecasting in power systems, load data is usually composed of multiple components with different time scales, such as long-term trends formed by seasonal variations, periodic fluctuations formed by diurnal electricity consumption patterns, and short-term random fluctuations caused by sudden electricity consumption behavior or disturbances in renewable energy output. These components correspond to energy distributions in different frequency ranges in the frequency domain, which can make load time series data exhibit significant non-stationarity and multi-scale coupling characteristics.

[0023] When using existing load forecasting models such as LSTM for load data prediction, they typically model the original load sequence holistically based on a single time scale. This makes it difficult to simultaneously account for both low-frequency trends and high-frequency fluctuations, often resulting in only a good fit to some features and impacting overall prediction accuracy. Furthermore, these load forecasting models mostly output deterministic results, lacking an effective representation of load fluctuation ranges and failing to reflect the uncertainties in actual operation. For example, under high-temperature conditions in summer, air conditioning electricity load fluctuates drastically in a short period and superimposed on the daily cycle curve, causing the load to exhibit significant abrupt changes and randomness, further increasing the difficulty of prediction. In this context, traditional single-time-scale load forecasting models not only struggle to accurately describe load change patterns, but their prediction results also fail to provide effective support for risk control and redundancy decisions in power equipment resource allocation.

[0024] Therefore, in this embodiment of the invention, starting from the intrinsic structure of historical load data sequences, a multi-scale feature is analyzed in depth, and an uncertainty characterization mechanism is introduced to improve the accuracy of prediction results while performing redundant evaluation of the reliability of prediction results, thereby improving the stability and reliability of subsequent optimization of power equipment resource allocation.

[0025] First, the historical load data sequence is transformed to the frequency domain using Fourier transform to obtain the corresponding spectral distribution map. When the energy in the spectral distribution map is mainly concentrated in a few frequency ranges, it indicates that the frequency structure of the load signal is relatively simple and the frequency richness is low. Conversely, when the energy in the spectral distribution map is distributed across multiple frequency intervals, and the amplitudes corresponding to multiple frequency components are all at high levels, it indicates that the load signal has obvious multi-scale superposition characteristics and its frequency richness is high. In this case, to more accurately reflect the changing characteristics at different scales, it is necessary to decompose the historical load data sequence into component signals of multiple different frequency scales to improve the accuracy and stability of subsequent prediction modeling. Therefore, the frequency richness of the spectral distribution map is evaluated to obtain the frequency richness evaluation value, which is used to characterize the complexity of the load signal in the frequency domain in the historical load data sequence and serves as the basis for subsequent multi-scale adaptive decomposition of the historical load data sequence. The method for evaluating the frequency richness of the spectral distribution map and obtaining the frequency richness evaluation value is as follows: The frequencies in the aforementioned spectral distribution map are linearly normalized to obtain normalized frequencies, ensuring that the normalized frequencies range from 0 to 1. This reduces the impact of differences in frequency distribution ranges among different load data on frequency richness assessment. The normalized frequencies are then paired to obtain multiple combinations. Based on the amplitude corresponding to each normalized frequency, the minimum amplitude in each combination is obtained. The frequency contribution weight of each combination is determined by its proportion among the minimum amplitudes of all combinations; a larger proportion corresponds to a larger frequency contribution weight.

[0026] In one embodiment, the formula for calculating the frequency contribution weight of the s-th combination is:

[0027] in, Let represent the contribution weights of normalized frequencies i and j in the s-th combination for measuring the frequency richness of load data, which are also the frequency contribution weights of the s-th combination. This represents the function that takes the minimum value, where n represents the number of combinations. , Let i and j represent the amplitudes of the normalized frequency i and j in the s-th combination, respectively, to reflect their energy proportion in the load signal.

[0028] It should be noted that, This represents the minimum amplitude among the amplitudes of the two normalized frequencies in the s-th combination. The larger the minimum amplitude, the higher the energy of both frequency components in the s-th combination, and the more significant their contribution to the frequency richness. This indicates that the minimum amplitude corresponding to the s-th combination is normalized.

[0029] Similarly, the frequency contribution weight of each combination is obtained, and then, combining the frequency contribution weight of each combination with each normalized frequency in the spectral distribution map, the frequency richness assessment value of the historical load data sequence is obtained: The absolute value of the difference between the two normalized frequencies contained in each combination is calculated respectively. Based on the frequency contribution weight of each combination, a weighted average of all absolute values ​​of difference is performed to obtain the frequency richness assessment value. The formula for calculating the frequency richness assessment value is as follows:

[0030] in, This represents the frequency richness assessment value of historical load data sequences. , Let i and j represent the values ​​of the normalized frequency i and j in the s-th combination, respectively. This represents the frequency contribution weight of the s-th combination, and n represents the number of combinations.

[0031] It should be noted that, This represents the absolute value of the difference between the normalized frequency i and the normalized frequency j in the s-th combination, used to reflect the degree of dispersion of the frequency distribution; This means that by weighted averaging the frequency differences under each combination, the overall frequency richness index, also known as the frequency richness assessment value, is obtained. It is used to indicate that the more dispersed the high-energy frequency components are in the spectrum distribution map, the more complex the frequency structure of the load data, and the higher its frequency richness, thus providing a more reliable parameter basis for subsequent multi-scale decomposition and prediction modeling.

[0032] Then, based on the frequency richness assessment value, the historical load data sequence is subjected to multi-scale adaptive decomposition. Since VMD (Variational Mode Decomposition) is mainly used to decompose complex non-stationary signals into multiple Intrinsic Mode Functions (IMFs), each IMF corresponds to the local features of different frequency components in the signal. The key parameter K in the VMD algorithm affects its decomposition performance. Therefore, in this embodiment of the invention, before performing multi-scale adaptive decomposition on the historical load data sequence, the number of adaptive components (i.e., the key parameter K) is obtained based on the frequency richness assessment value. This allows the decomposition result to dynamically match the characteristics of the historical load data sequence. The method for obtaining the number of adaptive components K is as follows: calculate the difference between the preset maximum and minimum number of components; multiply the difference by the frequency richness assessment value as the adjustment amount; and round down the sum of the minimum number of components and the adjustment amount to obtain the number of adaptive components when decomposing the historical load data sequence using the VMD algorithm.

[0033] In one embodiment, the formula for calculating the number of adaptive components is:

[0034] Where K represents the number of adaptive components when decomposing the historical load data sequence using the VMD algorithm. , These represent the preset minimum and maximum number of components, respectively. This represents the frequency richness assessment value of historical load data sequences. This indicates the floor function, used to ensure that the number of adaptive components K is an integer.

[0035] It should be noted that the settings This is used to ensure the most basic multi-scale expressive capabilities (corresponding to trend, period, and fluctuation terms), and to set... The maximum and minimum number of components (K) are used to limit the upper limit of the decomposition granularity to avoid noise amplification and increased computational complexity caused by excessive decomposition. The values ​​of the maximum and minimum number of components can be adjusted according to the specific application scenario and are not limited here. The value of Q ranges from 0 to 1. By introducing it as an adjustment parameter into the above mapping relationship, the decomposition scale can be adaptively adjusted. When the value of Q is large, it indicates that the frequency structure of the historical load data sequence is complex and has obvious multi-scale characteristics. At this time, the number of components K is increased to improve the decomposition fineness and fully extract the features of each scale. When the value of Q is small, it indicates that the structure of the historical load data sequence is relatively simple. Therefore, the number of components K is reduced accordingly to avoid excessive decomposition and improve computational efficiency and stability.

[0036] After determining the value of K, it is input into the VMD algorithm to decompose the historical load data sequence, resulting in K component signals with different center frequencies and bandwidth characteristics. Each component signal corresponds to the variation characteristics at different frequency scales: low-frequency components reflect long-term trends, mid-frequency components reflect periodic fluctuations, and high-frequency components reflect short-term disturbances. Through this decomposition process, the complex load signal is transformed into multiple clearly structured sub-signals, providing a foundation for subsequent component-level prediction modeling and uncertainty analysis, thereby improving the overall prediction accuracy and stability.

[0037] Step S102: For any component signal, set a prediction analysis interval for any component signal. Based on the difference between the load acquisition value and the load prediction value of each data point within the prediction analysis interval, perform prediction error analysis on any component signal to obtain the degree of error deviation under different preset quantiles. Combine the degree of error deviation under different preset quantiles to obtain the overall error value. Based on the energy intensity within the prediction analysis interval, obtain the prediction influence weight of any component signal.

[0038] Historical load data sequences are divided into K component signals with different frequency characteristics, enabling multi-scale representation of complex load fluctuations. Furthermore, LSTM load forecasting models are constructed for each component signal, and independent prediction error analysis is performed. Taking one component signal as an example, denoted as the i-th component signal, considering that component signals often have insufficient samples in the initial stage, the last m% of data points in the i-th component signal are set as the prediction analysis interval for the i-th component signal, where m is a positive integer, preferably set to 30, that is, the last 30% of data points in the i-th component signal are used as the prediction analysis interval. This reduces the interference of unstable data in the early stages on model training, thereby improving the stability and reliability of component-level prediction results.

[0039] Using a load forecasting model based on the i-th component signal, the load forecast value for each data point within the prediction analysis interval of the i-th component signal is obtained. This method is existing technology and will not be elaborated upon here. In actual power system operation scenarios, load changes are often influenced by multiple uncertainties, such as equipment start-up and shutdown behavior, changes in user-side electricity consumption patterns, and fluctuations in renewable energy output. Different frequency scale components exhibit significant differences in fluctuation amplitude, rate of change, and random disturbance characteristics, leading to inconsistencies in the numerical range and distribution of prediction errors for each component signal. If a uniform processing method is used for the prediction errors of each component signal, it is difficult to accurately reflect the uncertainty structure in the overall load prediction results. Therefore, different quantiles are set. Quantiles (such as 1%, 2%...99%) are statistical indicators that describe the location of data distribution. In prediction error analysis, quantiles are used to characterize the "typical value" and "degree of deviation" of the error. Their value range is (0, 1). Quantiles are usually expressed as decimals, such as 0.01, 0.05, 0.25, 0.5, and 0.95. Among them, 0.01–0.99 almost covers all common error distributions; 0.05–0.95 focuses on the main error and excludes extreme anomalies; 0.25 / 0.5 / 0.75 are quartiles, which describe the distribution pattern. Preferably, in this embodiment of the invention, the quantile is set to a value range of 0.01-0.99 for prediction error. Then, based on the difference between the load acquisition value and the load prediction value of each data point within the prediction analysis interval, prediction error analysis is performed on the i-th component signal to obtain the degree of error deviation under different preset quantiles, including: Calculate the absolute value of the difference between the load acquisition value and the load forecast value for each data point within the prediction analysis interval, and denot it as the prediction error for the corresponding data point, thereby obtaining a prediction error sequence. Based on the maximum and minimum values ​​in the prediction error sequence, obtain the theoretical error value corresponding to each preset quantile, where the theoretical error value corresponding to the r-th preset quantile is... , This represents the minimum value in the prediction error sequence. This represents the maximum value in the prediction error sequence. This represents the value of the r-th preset quantile.

[0040] For the r-th preset quantile, calculate the absolute value of the difference between the theoretical error value corresponding to the r-th preset quantile and each prediction error in the prediction error sequence, and obtain the average absolute value of the difference, which is denoted as the error deviation degree at the r-th preset quantile. The formula for calculating the error deviation degree at the r-th preset quantile is:

[0041] in, This represents the degree of error deviation at the r-th preset quantile, A represents the number of prediction errors in the prediction error sequence, and | represents the absolute value sign. This represents the a-th prediction error in the prediction error sequence.

[0042] Similarly, the error deviation degree under different preset quantiles is obtained. Based on the error deviation degree under each preset quantile, a weight is assigned to the theoretical error value corresponding to each preset quantile. The greater the error deviation degree, the smaller the weight. The theoretical error values ​​under all preset quantiles are then weighted and averaged to obtain the overall error value of the i-th component signal. The formula for calculating the overall error value is as follows:

[0043] in, This represents the overall error value of the i-th component signal. This indicates the preset number of quantiles. This represents the theoretical error value at the r-th preset quantile, where 1 represents a constant. This indicates the degree of error deviation at the r-th preset quantile. This represents the weight of the theoretical error value corresponding to the r-th preset quantile.

[0044] By weighting the theoretical error values ​​at each preset quantile based on the degree of error deviation, the quantiles that are closer to the actual error distribution occupy a larger proportion in the overall error estimation, thereby constructing an error estimate that is closer to the real situation and effectively improving the representativeness and credibility of the error assessment results. Furthermore, considering the different contributions of different component signals to the historical load data sequence, to avoid the excessive influence of low-energy components on the overall result, an influence weight based on the energy intensity of any component signal is introduced to obtain its prediction result on the final prediction result. Therefore, in this embodiment of the invention, the prediction influence weight of the i-th component signal is obtained according to the energy intensity within the prediction analysis interval of the i-th component signal. The specific method for obtaining this weight is as follows: Based on the load acquisition value of each data point within the prediction analysis interval, the accumulated load acquisition value is obtained and denoted as the energy intensity of the prediction analysis interval for the i-th component signal. The prediction influence weight of the i-th component signal is obtained based on its proportion among the energy intensities of all component signal prediction analysis intervals; a larger proportion results in a larger prediction influence weight. The formula for calculating the prediction influence weight of the i-th component signal is as follows:

[0045] in, Let represent the prediction influence weight of the i-th component signal, and N represent the number of data points within the prediction analysis interval of the i-th component signal. This represents the load acquisition value of the x-th data point within the predictive analysis interval of the i-th component signal. This represents the energy intensity of the prediction analysis interval for the i-th component signal. The energy intensity of the i-th component signal is weighted and normalized.

[0046] By introducing energy intensity as a weighting criterion, the contribution of prediction results of different component signals to the final load prediction can be quantified, allowing component signals with higher energy intensity and more significant impact on the load to play a greater role in the fusion process. This avoids interference from low-energy noise components and improves the accuracy of the final prediction interval estimation.

[0047] Step S103: Obtain the load forecast value corresponding to each component signal, and combine the overall error value and prediction influence weight of each component signal to obtain the load forecast value range after the current time, and allocate power equipment resources according to the load forecast value range.

[0048] Based on the above steps, the overall error value and prediction influence weight of each component signal can be obtained. Then, combined with the load prediction value of each component signal, the upper and lower limits of the load prediction value of the next time moment are constructed to form the load prediction value interval. Specifically: for any component signal, the any component signal is used as the input of the trained load prediction model, and the corresponding output is the load prediction value after the current time moment; the sum of the load prediction value and the overall error value of the any component signal is calculated and recorded as the upper adjustment load prediction value of the any component signal; the difference between the load prediction value and the overall error value of the any component signal is calculated and recorded as the lower adjustment load prediction value of the any component signal. Based on the prediction influence weight of each component signal, the upper-adjustment load prediction values ​​of all component signals are weighted and averaged to obtain the upper limit of the load prediction after the current time; the lower-adjustment load prediction values ​​of all component signals are weighted and averaged to obtain the lower limit of the load prediction after the current time; the upper limit of the load prediction after the current time and the lower limit of the load prediction after the current time constitute the load prediction value interval after the current time.

[0049] The formula for calculating the load forecast interval after the current time is as follows:

[0050]

[0051] in, This represents the upper limit of the load forecast range. This represents the lower limit of the load forecast range, and K represents the number of component signals. This represents the load forecast value based on the i-th component signal, which is the load forecast value for the next time step obtained from the current time step using the i-th component signal. This represents the prediction influence weight of the i-th component signal. This represents the overall error value of the i-th component signal.

[0052] Let the range of load forecast values ​​after the current time be denoted as . By accurately estimating the load forecast range, a more robust decision-making basis can be provided for the power system in equipment start-up and shutdown scheduling, load allocation, and resource allocation. This enables scheduling strategies to simultaneously consider safety margins and economic efficiency, effectively reducing resource redundancy or supply shortages caused by forecast deviations, and improving the overall stability and reliability of the system. Therefore, obtaining the load forecast range... Optionally, the upper limit of the load forecast range can be used as the safety constraint boundary for power system operation to verify the power generation capacity, substation capacity, and transmission and distribution capacity, ensuring sufficient supply capacity under high load conditions. Next, the median value of the load forecast range can be used as the benchmark for normal operation. Based on this, the unit combination and load allocation can be determined using existing economic dispatch methods. Furthermore, the normal operation benchmark can be dynamically adjusted according to the actual load redundancy. For example, when system redundancy is low, a certain proportion of load allocation can be appropriately increased based on the median benchmark to enhance supply capacity and ensure that redundancy is not lower than a preset threshold (e.g., 20%, which is just an example and can be adjusted according to actual conditions). Conversely, when redundancy is high, the load allocation ratio can be appropriately reduced to improve resource utilization efficiency. Through these measures, dynamic coordination and continuous optimization of power equipment resource allocation between safety and economy can be achieved while ensuring supply and demand balance.

[0053] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for optimizing power equipment resource allocation based on intelligent analysis, characterized in that, The method includes: Load data is collected from the power system to obtain a historical load data sequence up to the current moment. The historical load data sequence is converted to the frequency domain to obtain a corresponding spectrum distribution map. The frequency richness of the spectrum distribution map is evaluated to obtain a frequency richness evaluation value. Based on the frequency richness evaluation value, the adaptive number of components is obtained when decomposing the historical load data sequence using the VMD algorithm, and the historical load data sequence is decomposed into multiple component signals. For any component signal, a prediction analysis interval is set for the any component signal. Based on the difference between the load acquisition value and the load prediction value of each data point within the prediction analysis interval, prediction error analysis is performed on the any component signal to obtain the error deviation degree under different preset quantiles. The overall error value is obtained by combining the error deviation degree under different preset quantiles. The prediction influence weight of the any component signal is obtained based on the energy intensity within the prediction analysis interval. The load forecast value corresponding to each component signal is obtained, and combined with the overall error value and prediction influence weight of each component signal, the load forecast value range after the current time is obtained, which is used for power equipment resource allocation optimization.

2. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The step of evaluating the frequency richness of the spectral distribution map to obtain a frequency richness evaluation value includes: The frequencies in the spectrum distribution map are linearly normalized to obtain normalized frequencies. The normalized frequencies are then combined in pairs. Based on the amplitude corresponding to each normalized frequency, the minimum amplitude in each combination is obtained. Based on the proportion of the minimum amplitude in each combination to the minimum amplitude in all combinations, the frequency contribution weight of each combination is obtained. The larger the proportion, the larger the corresponding frequency contribution weight. Calculate the absolute value of the difference between the two normalized frequencies contained in each combination, and perform a weighted average of all the absolute values ​​of the difference based on the frequency contribution weight of each combination to obtain the frequency richness evaluation value.

3. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The step of obtaining the adaptive number of components when decomposing the historical load data sequence using the VMD algorithm based on the frequency richness evaluation value includes: Calculate the difference between the preset maximum number of components and the minimum number of components. Use the product of the difference and the frequency richness evaluation value as the quantity adjustment amount. Round down the sum of the minimum number of components and the quantity adjustment amount to obtain the adaptive number of components when decomposing the historical load data sequence using the VMD algorithm.

4. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The step of performing prediction error analysis on any component signal based on the difference between the load acquisition value and the load prediction value at each data point within the prediction analysis interval, to obtain the degree of error deviation under different preset quantiles, includes: Calculate the absolute value of the difference between the load collection value and the load prediction value for each data point within the prediction analysis interval, and record it as the prediction error for the corresponding data point, thereby obtaining the prediction error sequence; based on the maximum and minimum values ​​in the prediction error sequence, obtain the theoretical error value corresponding to each preset quantile; For any preset quantile, calculate the absolute value of the difference between the theoretical error value corresponding to the preset quantile and each prediction error in the prediction error sequence, and obtain the average absolute value of the difference, which is denoted as the error deviation degree under the preset quantile.

5. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 4, characterized in that, The process of obtaining the overall error value by combining the error deviation levels under different preset quantiles includes: Based on the degree of error deviation at each preset quantile, a weight is assigned to the theoretical error value corresponding to each preset quantile. The greater the degree of error deviation, the smaller the weight. The theoretical error values ​​at all preset quantiles are then weighted and averaged to obtain the overall error value.

6. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The step of obtaining the prediction influence weight of any component signal based on the energy intensity within the prediction analysis interval includes: Based on the load collection value of each data point within the prediction analysis interval, the accumulated load collection value is obtained and recorded as the energy intensity of the prediction analysis interval of any component signal. Based on the proportion of the energy intensity of the prediction analysis interval of any component signal in the energy intensity of all component signals' prediction analysis intervals, the prediction influence weight of any component signal is obtained. The larger the proportion, the greater the corresponding prediction influence weight.

7. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The step of obtaining the load forecast value corresponding to each component signal, and combining the overall error value and prediction influence weight of each component signal to obtain the load forecast value range after the current time, includes: For any component signal, the any component signal is used as the input of the trained load forecasting model, and the corresponding output is the load forecast value after the current time. The sum of the load forecast value and the overall error value of the any component signal is calculated and recorded as the upward adjustment load forecast value of the any component signal. The subtraction between the load forecast value and the overall error value of the any component signal is calculated and recorded as the downward adjustment load forecast value of the any component signal. Based on the prediction influence weight of each component signal, the upper adjustment load prediction value of all component signals is weighted and averaged to obtain the upper limit of the load prediction after the current time; the lower adjustment load prediction value of all component signals is weighted and averaged to obtain the lower limit of the load prediction after the current time. The load forecast range after the current time is formed by the upper limit and lower limit of the load forecast after the current time.

8. The method for optimizing power equipment resource allocation based on intelligent analysis according to claim 1, characterized in that, The prediction analysis interval of any component signal refers to the last m% of data points in any component signal, where m is a positive integer.

Images