A method and system for modeling electricity load based on time-of-use electricity consumption

Abstract

This application relates to the field of electricity load analysis technology, and discloses a method and system for modeling electricity load based on time-of-use (TOU) electricity consumption. The method includes: based on TOU electricity consumption data collection from smart meters and equipment start-up / stop frequency identification, extracting the basic load characteristics during peak, valley, and normal periods using a TOU empirical mode decomposition algorithm; obtaining the equipment start-up / stop load correlation characteristics by combining a TOU variational mode decomposition algorithm with the equipment start-up / stop state matrix; using a TOU hyperparameter optimization algorithm to predict the load for each time period; and finally establishing a start-up / stop frequency influence weighting mechanism through a TOU load state fusion algorithm to obtain the TOU electricity load modeling result. This application addresses the problems of existing TOU electricity load modeling methods, such as lack of equipment start-up / stop status awareness, uneven modeling accuracy between time periods, and poor continuity of time period switching.

Description

Technical Field

[0001] This application relates to the field of electricity load analysis technology, and in particular to a modeling method and system for electricity load models based on time-of-use electricity consumption. Background Technology

[0002] Currently, time-of-use (TOU) electricity load modeling methods are widely used in power system planning, demand response, and smart grid management. Existing technologies mainly rely on historical TOU electricity data for modeling, using time series analysis methods such as ARIMA models, regression analysis, or machine learning algorithms such as neural networks to predict load. These methods divide a 24-hour day into different time periods, collect actual electricity consumption data for each period, and then use this historical TOU electricity data to identify electricity consumption patterns and build mathematical models, thereby achieving the prediction of future electricity load. Traditional modeling methods typically employ a uniform algorithm framework when processing TOU electricity data, applying the same processing strategies to data from different time periods.

[0003] However, existing technologies have significant technical limitations. First, traditional methods primarily rely on historical time-of-use electricity data for modeling, failing to incorporate the operational characteristics of major household appliances as key input variables. This results in models that cannot effectively identify and quantify the specific impact and patterns of high-power appliance start-up and shutdown behavior on time-of-use load fluctuations. Second, existing static modeling methods struggle to accurately capture the cumulative effects and interrelationships of electricity demand when multiple high-power appliances operate simultaneously or alternately, particularly the dynamic coupling relationships during peak electricity consumption periods. Furthermore, when residents experience seasonal changes in their electricity usage habits, adjustments to their daily routines, or the addition of new appliances, existing models lack a mechanism for quickly identifying changes and adaptively adjusting parameters, leading to a significant decrease in prediction accuracy.

[0004] Due to the lack of real-time sensing capabilities for the start-up and shutdown status of home appliances, traditional modeling methods cannot establish a precise correlation between the start-up and shutdown behavior of equipment and time-of-use load fluctuations. This leads to significant differences in the accuracy of load forecasting at different time periods (peak, flat, and valley). In particular, when the start-up and shutdown frequency of equipment varies greatly between time periods, a single modeling strategy is difficult to meet the accuracy requirements of each time period simultaneously. Furthermore, the results of independent forecasts for each time period often exhibit load jumps and logical inconsistencies at time period switching points, which seriously affects the overall modeling accuracy and practicality of the time-of-use electricity load model. Summary of the Invention

[0005] This application provides a method and system for modeling electricity load based on time-of-use electricity consumption, which solves the problems of lack of equipment start-up and shutdown status perception, uneven modeling accuracy between time periods, and poor continuity of time period switching in existing time-of-use electricity load modeling methods.

[0006] Firstly, this application provides a method for modeling an electricity load model based on time-of-use (TOU) electricity consumption. The method includes: collecting TOU data from residential users using smart meters to obtain TOU sequences; simultaneously identifying the start-stop frequency of household appliances using an equipment monitoring module to obtain an equipment start-stop state matrix; based on the TOU sequences, using a TOU empirical mode decomposition algorithm to perform differentiated decomposition processing on the data for each time period to obtain peak-valley-normal period basic load characteristics; using a TOU variational mode decomposition algorithm to identify start-stop fluctuation signals from the basic load characteristics, and combining this with the equipment start-stop state matrix to obtain equipment start-stop load correlation characteristics; based on the equipment start-stop load correlation characteristics, using a TOU hyperparameter optimization algorithm to perform load prediction processing on the start-stop frequency for each time period to obtain time period start-stop load prediction results; and integrating the time period start-stop load prediction results using a TOU load state fusion algorithm to establish a start-stop frequency influence weighting mechanism to obtain TOU electricity load modeling results.

[0007] Secondly, this application provides a modeling system for electricity load based on time-of-use electricity consumption, the modeling system comprising:

[0008] The data acquisition module is used to collect time-of-use electricity consumption data from residential users through smart meters to obtain time-of-use electricity consumption sequences. At the same time, the device monitoring module identifies the start-stop frequency of household appliances to obtain a device start-stop status matrix.

[0009] The decomposition module is used to perform differential decomposition processing on the data of each time period based on the time-sharing power consumption sequence using the time-sharing power consumption empirical mode decomposition algorithm to obtain the basic load characteristics of the peak-valley and normal periods.

[0010] The identification module is used to process the start-stop fluctuation signal identification of the basic load characteristics through the time-division variational mode decomposition algorithm, and combine it with the equipment start-stop state matrix to obtain the equipment start-stop load correlation characteristics.

[0011] The prediction module is used to perform load prediction processing on the start-up and stop frequencies of each time period based on the start-up and stop load correlation characteristics of the equipment, and to obtain the start-up and stop load prediction results for each time period.

[0012] The integration module is used to integrate the start-stop load prediction results of the time period through the time-sharing load status fusion algorithm, establish the start-stop frequency influence weight mechanism, and obtain the time-sharing power load modeling results.

[0013] Thirdly, a time-of-use (TOU) power load modeling device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the TOU power load modeling device to execute the above-described TOU power load modeling method.

[0014] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to execute the above-described method for modeling the electricity load model based on time-of-use electricity.

[0015] The technical solution provided in this application combines time-of-use electricity collection by smart meters with start-stop frequency identification by equipment monitoring modules, establishing a direct correlation between time-of-use electricity data and the operating status of home appliances. This solves the technical deficiency of traditional methods that rely solely on historical electricity data and cannot identify the impact of equipment behavior. The differentiated decomposition processing of the time-of-use electricity empirical mode decomposition algorithm adaptively sets parameters to address the differences in load characteristics during peak, valley, and normal periods, effectively extracting the basic load characteristics of each time period and laying the foundation for subsequent accurate modeling. The time-of-use variational mode decomposition algorithm, by combining the equipment start-stop state matrix to identify start-stop fluctuation signals, establishes a precise mapping relationship between equipment start-stop behavior and load changes, significantly enhancing the model's ability to perceive changes in equipment behavior. The time-of-use hyperparameter optimization algorithm constructs differentiated network structures and performs independent parameter optimization for the start-stop frequency characteristics of each time period, avoiding the problem that a single model cannot simultaneously adapt to the differences in complexity across time periods, and greatly improving the accuracy and stability of load prediction for each time period. The time-of-use load fusion algorithm establishes a weighting mechanism to integrate the prediction results of each time period by setting up a start-stop frequency influence mechanism. This maintains the load continuity between time periods and ensures the logical consistency between the fusion results and the actual start-stop behavior of the equipment, thereby obtaining high-precision and physically reasonable time-of-use power load modeling results.

[0016] The time-of-use (TOU) power consumption empirical mode decomposition algorithm, through adaptive noise enhancement and differentiated mode settings, can accurately separate transient load changes caused by the start-up and shutdown of household appliances from the basic power load. This separation capability is of great significance for understanding the patterns of residential electricity consumption behavior. The TOU variational mode decomposition algorithm, combined with an innovative design of equipment state constraints, enables the algorithm to accurately identify the independent contribution of each device to the load in complex multi-device concurrent start-up and shutdown environments. This plays a crucial role in achieving refined power management and demand response. The TOU hyperparameter optimization algorithm, based on differentiated network configuration strategies for equipment usage patterns in different time periods, fully utilizes the inherent characteristics of complex equipment start-up and shutdown during peak hours, relatively stable during normal periods, and simple during valley periods. Through reasonable matching of deep, medium, and shallow networks, it significantly reduces computational complexity while ensuring prediction accuracy. The equipment state transfer matrix and continuity constraint mechanism in the TOU load state fusion algorithm effectively solve the problem of smooth transition of residential electricity consumption behavior during time period switching. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of an embodiment of the electricity load modeling method based on time-of-use electricity consumption in this application.

[0019] Figure 2 This is a schematic diagram of an embodiment of the power load modeling system based on time-of-use electricity consumption in this application.

[0020] Figure 3 This is a schematic block diagram of the structure of the power load modeling device based on time-of-use electricity in an embodiment of the present invention. Detailed Implementation

[0021] This application provides a method and system for modeling an electricity load model based on time-of-use electricity consumption. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application 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 described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0022] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the electricity load modeling method based on time-of-use electricity consumption in this application includes:

[0023] Step S101: Collect time-of-use electricity consumption data from residential users using smart meters to obtain time-of-use electricity consumption sequences. At the same time, identify the start-stop frequency of household appliances using the device monitoring module to obtain the device start-stop status matrix.

[0024] Step S102: Based on the time-of-use electricity sequence, the time-of-use electricity empirical mode decomposition algorithm is used to perform differential decomposition processing on the data of each time period to obtain the basic load characteristics of the peak-valley and normal periods.

[0025] Step S103: The basic load characteristics are processed by the time-division variational mode decomposition algorithm to identify and process start-stop fluctuation signals. Combined with the equipment start-stop state matrix, the equipment start-stop load correlation characteristics are obtained.

[0026] Step S104: Based on the equipment start-up and shutdown load correlation characteristics, the time-sharing hyperparameter optimization algorithm is used to perform load prediction processing on the start-up and shutdown frequencies of each time period to obtain the start-up and shutdown load prediction results for each time period.

[0027] Step S105: Integrate the start-stop load prediction results of the time period through the time-sharing load status fusion algorithm, establish the start-stop frequency influence weight mechanism, and obtain the time-sharing power load modeling results.

[0028] It is understood that the executing entity of this application can be a power load modeling system based on time-of-use electricity consumption, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0029] Specifically, the 24-hour time period is divided into three intervals—peak, normal, and valley—according to the time-of-use pricing policy. Smart meters sample residential users' electricity consumption at 15-minute intervals, forming 96 time-of-use electricity data points. These data points are arranged and combined according to time series to form a time-of-use electricity sequence that includes peak, valley, and flat load variations. Simultaneously, a target monitoring device list is established by setting power thresholds to filter high-power household appliances such as air conditioners, water heaters, and washing machines. Current sensors monitor the start-up and stop-down status of each device in the list, recording the number of times the device starts and stops. The monitoring data is organized into an device start-up and stop-down status matrix according to device number and timestamp.

[0030] Differential noise parameters for each time period are calculated based on the load fluctuation variance during peak, normal, and valley periods. Noise amplitude is adaptively adjusted according to the load fluctuation characteristics of each time period, and paired white noise is added to the corresponding time period's electricity data for noise enhancement. During empirical mode decomposition, a number of high-frequency modes are set during peak periods to capture drastic load fluctuations, a number of medium-frequency modes are set during normal periods to extract stable load characteristics, and a number of low-frequency modes are set during valley periods to analyze the base load change trend. The inherent mode components of each time period are integrated and averaged to eliminate noise influence, retaining the true load change trend to form the cleaned mode components for each time period. Based on the frequency characteristics of the cleaned mode components, base load modes are screened to extract mode components reflecting the basic electricity consumption patterns of each time period.

[0031] High-frequency modal components are extracted from the basic load characteristics during peak, valley, and normal periods. These high-frequency components contain transient load fluctuation information caused by equipment start-up and shutdown. Differentiated penalty parameters are set according to the differences in load characteristics during peak, normal, and valley periods to construct time-period differentiated decomposition parameters. Equipment state constraints are established based on the equipment start-up and shutdown state matrix, and the equipment start-up and shutdown states are correlated with load fluctuation signals. The time-sharing variational modal decomposition algorithm takes the high-frequency load fluctuation signals and equipment state constraints of each time period as input, and performs matching and quantification of equipment start-up and shutdown times with load fluctuation peak times through constraint weight coefficient calculation. The algorithm uses an iterative decomposition strategy to extract variational modes, separates each variational mode by combining frequency domain filtering and time domain constraints, and adjusts the center frequency and bandwidth parameters of each mode according to the equipment state constraints.

[0032] The time-sharing hyperparameter optimization algorithm separates the equipment start-stop load correlation characteristics based on the differences in start-stop frequencies during peak, normal, and valley periods, extracting equipment start-stop load correlation data for each period. The algorithm constructs differentiated network structures for different periods: a deep network structure is used during peak periods to handle complex load changes; a mid-level network structure is used during normal periods to handle stable load patterns; and a shallow network structure is used during valley periods to analyze simple load patterns. The network for each period undergoes hyperparameter optimization using a successive halving strategy, independently optimizing the learning rate, number of neurons, and dropout rate. The optimized network uses historical start-stop frequency data and corresponding load data for parameter learning, and the current start-stop frequency data is input into the trained network for load prediction calculation, obtaining the load prediction values ​​for each period.

[0033] The time-of-use load fusion algorithm calculates load continuity constraints at the switching points of peak, normal, and valley periods in the time-based start-stop load forecast results, establishing load continuity constraints between adjacent time periods. It constructs an equipment state transfer weight matrix based on the equipment start-stop state matrix, calculating the weight of equipment state influence between adjacent time periods. The fusion algorithm calculates the weighted fusion weight of the forecast results for each time period based on the time-based switching continuity constraint parameters and the equipment state transfer weight matrix, assigning time-related fusion weight coefficients to each time period's forecast results. The time-based start-stop load forecast results are weighted and summed according to the fusion weight coefficients, and a comprehensive calculation is performed combining the weight coefficients of the equipment start-stop frequency's impact on the load. The fused load forecast sequence is verified for logical consistency with the actual equipment start-stop states through equipment state consistency checks, and the time-of-use electricity load modeling results are formed through iterative adjustments and optimizations.

[0034] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0035] The 24-hour time period is divided into peak, normal and off-peak periods to obtain three time-of-use electricity price intervals.

[0036] Based on three time-of-use electricity price periods, the electricity data collected by smart meters is sampled at 15-minute intervals to obtain the time-of-use electricity data points for each period.

[0037] The time-of-use electricity data points for each time period are arranged and combined according to the time series to obtain the time-of-use electricity series that includes the changes in peak, valley and flat loads;

[0038] Based on the power threshold of household appliances, air conditioners, water heaters, and washing machines are screened for high-power equipment to obtain a list of target monitoring equipment.

[0039] Based on the target monitoring equipment list, the start-stop status of each device is monitored by current sensors, and the number of times the device is started and stopped is recorded to obtain the device start-stop status matrix.

[0040] Specifically, the time-of-use pricing policy divides 24 hours into three periods: peak hours, off-peak hours, and valley hours. Peak hours cover peak electricity consumption periods, normal hours cover daily electricity consumption, and valley hours correspond to the nighttime off-peak electricity consumption period. This division reflects the temporal regularity of residential electricity consumption behavior and corresponds to different electricity price levels, forming a tiered structure where peak hour electricity prices are higher than normal hour prices, and normal hour prices are higher than valley hour prices.

[0041] The 15-minute interval sampling process segments the smart meter's electricity consumption data into fixed time windows, with each window constituting a sampling point. The smart meter uses its built-in metering chip to monitor current and voltage in real time, calculates the instantaneous power value, and then integrates this power value over time to obtain the cumulative electricity consumption over 15 minutes. This sampling method ensures the temporal continuity and integrity of the data while balancing data accuracy and storage space requirements.

[0042] The permutation and combination processing concatenates the time-of-use electricity data points from different time periods according to their timestamp order, forming a two-dimensional data sequence containing both time and electricity attributes. Each data point includes a time identifier and a corresponding electricity value. The time identifier indicates the time period type and specific time to which the data point belongs, while the electricity value records the electricity consumption within that time window. Logical connections are established between data points based on their temporal sequence, forming a continuous time-of-use electricity sequence. This sequence includes load variation information for peak, valley, and flat periods.

[0043] The high-power device screening process sets power thresholds based on the rated power parameters of household appliances, typically set at 1000 watts. Air conditioners generally have rated power between 1500 and 3000 watts, water heaters between 2000 and 4000 watts, and washing machines between 500 and 1200 watts. The screening algorithm iterates through the power parameters of all appliances in the household, adding devices with rated power exceeding the threshold to the target monitoring device list, thus forming a list of high-power devices requiring focused monitoring.

[0044] Start-up and shutdown status monitoring is achieved through current sensors installed on the power lines of each device. These current sensors detect current changes in the circuit using the Hall effect or electromagnetic induction principle. When the device starts, the current jumps from zero to the operating current; when the device stops, the current drops from the operating current to zero. The monitoring algorithm determines the start-up and shutdown status of the device by setting current thresholds. A start event is recorded when the current value exceeds the start threshold, and a stop event is recorded when the current value is below the stop threshold. The start-up count algorithm calculates the number of times the device transitions from the off state to the on state within the monitoring period, and the stop count algorithm calculates the number of times the device transitions from the on state to the off state. The device start-up and shutdown status matrix organizes the start-up and shutdown count data of each device according to device number and time period. Rows in the matrix represent different devices, columns represent different time periods, and matrix element values ​​represent the start-up and shutdown counts of the corresponding device within the corresponding time period.

[0045] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0046] Based on the load fluctuation variance of peak, normal, and valley periods, adaptive noise amplitude calculation is performed on the time-of-use electricity sequence to obtain differentiated noise parameters for each time period.

[0047] Based on the differentiated noise parameters for each time period, pairs of white noise are added to the corresponding time period's electricity data for noise enhancement, resulting in a noisy time period electricity sequence.

[0048] The noisy time-period electricity sequence is processed by empirical mode decomposition according to the number of high-frequency modes during peak periods, the number of mid-frequency modes during normal periods, and the number of low-frequency modes during valley periods to obtain the inherent mode components of each time period;

[0049] The inherent modal components of each time period are integrated and averaged to eliminate the influence of noise and retain the true load change trend, thus obtaining the time period purification modal components;

[0050] Based on the frequency characteristics of the purification mode components during different time periods, the basic load mode screening process is performed to extract the mode components that reflect the basic electricity consumption patterns during each time period, thereby obtaining the basic load characteristics during peak, valley, and normal periods.

[0051] Specifically, the adaptive noise amplitude calculation process first calculates the load fluctuation variance of the time-of-use electricity data during peak, normal, and valley periods. This variance reflects the dispersion and volatility of the electricity data in each period. Peak periods experience greater load fluctuations due to frequent start-ups and shutdowns of appliances, resulting in higher calculated variance values. Normal periods show relatively stable load fluctuations with moderate variances, while valley periods exhibit the lowest variance due to minimal load fluctuations. The algorithm sets differentiated noise parameters based on the variance values ​​for each period. The noise amplitude is directly proportional to the load fluctuation variance of the corresponding period; periods with larger fluctuation variances have higher noise amplitudes, while those with smaller fluctuation variances have lower noise amplitudes. The noise enhancement process employs paired white noise addition technology. Paired white noise refers to two sets of random noise signals with the same amplitude but opposite signs. The algorithm adds positive and negative noise to the electricity data for the corresponding periods, creating two data copies containing noise. Positive noise increases the original electricity value, while negative noise decreases it. The two sets of noise have the same statistical characteristics but opposite polarities. This paired addition method ensures that the effects of noise on the data cancel each other out statistically, while increasing the randomness and diversity of the data. The noisy time-period electricity sequence contains both the original electricity information and the added noise information. The noise component helps the Empirical Mode Decomposition (EMD) algorithm to better identify the true modal characteristics in the data.

[0052] Empirical Mode Decomposition (EMD) sets different modal quantity parameters based on the differences in load characteristics across different time periods. During peak periods, a larger number of high-frequency modes are set, typically 8 to 10 modal components, to capture high-frequency load fluctuations caused by frequent start-stop cycles of high-power equipment such as air conditioners and water heaters. During normal periods, a larger number of mid-frequency modes are set, typically 5 to 7 modal components, to analyze the periodic operating patterns of equipment such as washing machines and microwave ovens. During off-peak periods, a smaller number of low-frequency modes are set, typically 3 to 5 modal components, primarily to analyze low-frequency load changes in continuously operating equipment such as refrigerators and routers. The decomposition process uses a filtering algorithm to extract intrinsic modal components layer by layer. Each modal component represents an oscillating component with specific frequency characteristics in the original signal. The algorithm sequentially separates each modal component from high to low frequencies from the noisy time-period power consumption sequence until the remaining components no longer contain obvious oscillating characteristics.

[0053] The integrated averaging process averages the intrinsic modal components of each time period obtained after adding different noises multiple times. Since paired white noise is added, the positive and negative noises cancel each other out in multiple calculations, preserving and enhancing the true load change trend. During the averaging process, the randomness of the noise components causes them to statistically tend towards zero, while the determinism of the true load signal strengthens them during averaging. The time-period cleaned modal components are the result of the integrated averaging, containing the true load change information after removing noise interference, while maintaining the differences in load characteristics across time periods.

[0054] The base load mode selection process selects modes based on the frequency characteristics of the purified mode components for each time period. The algorithm calculates the dominant frequency and energy distribution of each mode component. The dominant frequency reflects the speed of mode oscillation, and the energy distribution reflects the mode's contribution to the original signal. Base load modes typically correspond to low-frequency, high-energy mode components, reflecting the basic patterns and long-term trends of residential electricity consumption. The selection algorithm sets frequency and energy thresholds; modes with frequencies below the thresholds and energy levels above the thresholds are identified as base load modes. The base load characteristics during peak, valley, and normal periods are reconstructed from the selected base load modes, containing fundamental information on electricity consumption patterns for each time period and removing interference from transient fluctuations such as equipment start-up and shutdown.

[0055] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0056] High-frequency modal components are extracted from the characteristics of the base load during peak, valley and normal periods, and start-stop fluctuation signals are filtered and processed to obtain high-frequency load fluctuation signals for each time period.

[0057] Based on the differences in load characteristics during peak, normal, and valley periods, differentiated penalty parameters are set for high-frequency load fluctuation signals in each time period to obtain time-period differentiated decomposition parameters.

[0058] Based on the equipment start-stop state matrix, the equipment state constraint conditions are constructed and the constraint mechanism is established. The equipment start-stop state is associated with the load fluctuation signal to obtain the equipment state constraint conditions.

[0059] The high-frequency load fluctuation signals and equipment status constraints of each time period are input into the variational mode decomposition algorithm for constraint variational solution processing. The optimal variational mode is obtained through iterative optimization to obtain the equipment-related variational mode components.

[0060] The start-stop behavior matching and verification process is performed on the equipment-related variational mode components to identify the load change modes corresponding to the start-stop times of home appliances, thereby obtaining the equipment start-stop load correlation characteristics.

[0061] Specifically, the high-frequency modal component extraction process identifies and separates high-frequency components from the base load characteristics during peak-valley and normal periods through frequency domain analysis. High-frequency modal components refer to high-frequency oscillation components, which typically correspond to transient load changes caused by the start-up and shutdown operations of household appliances. The algorithm uses spectrum analysis to calculate the frequency distribution of the base load characteristics, sets frequency thresholds to classify modal components into high-frequency and low-frequency categories, and classifies modes with frequencies exceeding the threshold as high-frequency modal components. The start-up and shutdown fluctuation signal screening process further identifies signal components related to equipment start-up and shutdown behavior from the high-frequency modal components. Time-frequency analysis is used to analyze the distribution characteristics of high-frequency modes in the time domain, and signal segments with significant energy peaks near the equipment start-up and shutdown times are selected as start-up and shutdown fluctuation signals.

[0062] Differentiated penalty parameter settings are implemented to assign different decomposition parameters to each time period based on the significant differences in load characteristics during peak, normal, and valley periods. The penalty parameter is a key parameter controlling the modal bandwidth in the variational mode decomposition algorithm. A larger parameter value results in a narrower modal bandwidth and higher decomposition accuracy, while a smaller parameter value results in a wider modal bandwidth and higher decomposition coarsening. During peak periods, due to frequent equipment start-ups and shutdowns and complex load fluctuations, the algorithm sets a larger penalty parameter value to obtain more refined modal decomposition results, accurately identifying the start-up and shutdown transient characteristics of high-power equipment such as air conditioners and water heaters. During normal periods, equipment operation is relatively stable, and the algorithm sets a moderate penalty parameter value to balance decomposition accuracy and computational efficiency. During valley periods, load fluctuations are minimal, and the algorithm sets a smaller penalty parameter value to focus on the slow changing trend of the base load. The time-period differentiated decomposition parameters include key parameters controlling the variational mode decomposition process, such as the penalty parameter value, tolerance parameter, and number of iterations for each time period.

[0063] The equipment state constraint construction process establishes a correlation constraint relationship between equipment operating status and load fluctuation signals based on the equipment start-stop state matrix. The equipment start-stop state matrix records the number of start-stop cycles and specific time information for each device in different time periods. The constraint construction algorithm extracts the equipment start-up and stop times as time constraint points. The correlation constraint mechanism requires that the modal components obtained from variational mode decomposition exhibit positive load surge characteristics near the equipment start-up time and negative load drop characteristics near the equipment stop time. The constraint mechanism establishment process integrates equipment state information into the objective function of variational mode decomposition through a weighting function, assigning higher weights to load changes corresponding to equipment start-stop times and lower weights to load changes outside of start-stop times. The equipment state constraints include three dimensions of restrictions: time constraints, amplitude constraints, and correlation constraints.

[0064] The constrained variational solution process takes high-frequency load fluctuation signals from different time periods and equipment state constraints as input data and feeds them into the variational mode decomposition algorithm. The variational mode decomposition algorithm decomposes the input signal into several variational mode components with different frequency characteristics by solving a constrained optimization problem. Each mode component has a compact spectral support and a clear physical meaning. The constrained variational solution process uses an alternating direction multiplier method for iterative optimization, with the algorithm alternately updating the variational modes and Lagrange multipliers until the convergence condition is met. During the iterative optimization process, the equipment state constraints, as additional constraints, affect the shape and frequency distribution of the mode components, making the decomposed mode components more correlated with equipment start-up and shutdown behavior. The equipment-associated variational mode components are the output of the constrained variational solution process, containing load change information highly correlated with specific equipment start-up and shutdown behavior.

[0065] The start-stop behavior matching verification process verifies the degree of matching between the device-associated variational mode components and the start-stop times of the home appliances through time alignment and correlation analysis. The algorithm calculates the energy distribution and peak position of each variational mode component and compares the time position of the mode energy peak with the start-stop times recorded in the device start-stop state matrix. The matching verification algorithm sets a time tolerance range; a successful match is considered achieved when the time difference between the mode energy peak time and the device start-stop time is less than the tolerance range. Correlation analysis calculates the cross-correlation coefficient between the variational mode components and the device start-stop state sequence; modes with correlation coefficients exceeding a threshold are considered to have a strong correlation with the corresponding device. The device start-stop load correlation features are composed of the variational mode components that have passed the matching verification; these features contain detailed information on the impact of home appliance start-stop behavior on time-of-use electricity load.

[0066] In one specific embodiment, the process of inputting the high-frequency load fluctuation signals of each time period and the equipment state constraints into the variational mode decomposition algorithm for constraint variational solution processing can specifically include the following steps:

[0067] Based on the equipment status constraints, the constraint weight coefficients of the high-frequency load fluctuation signals in each time period are calculated and processed. The equipment start-up and shutdown times are matched and quantified with the peak load fluctuation times to obtain the constraint weight coefficients corresponding to the start-up and shutdown states of each equipment.

[0068] Based on the constraint weight coefficient, the high-frequency load fluctuation signals of each time period are processed by weighted constraint variational objective construction, and the load fluctuation signals corresponding to the equipment start-up and shutdown times are given higher weights to obtain the equipment constraint variational optimization objective.

[0069] An iterative decomposition strategy is used to extract variational modes for the equipment-constrained variational optimization objective. Each variational mode is separated by a combination of frequency domain filtering and time domain constraints to obtain the initial variational mode components.

[0070] The initial variational mode components are processed by iterative optimization algorithm to adjust the mode parameters. The center frequency and bandwidth parameters of each mode are continuously adjusted according to the equipment state constraints to obtain the optimized variational mode components.

[0071] The correlation degree of the optimized variational mode components at the equipment start-up and shutdown times is verified. The correlation between each mode and the time nodes of the equipment start-up and shutdown state matrix is ​​calculated, and high correlation modes are selected to obtain the equipment-related variational mode components.

[0072] Specifically, the constraint weight coefficient calculation process uses a time alignment algorithm to precisely match the equipment start-up and shutdown times with the peak times of load fluctuations. The algorithm first identifies the peak and trough points of load fluctuations from the high-frequency load fluctuation signals of each time period. Peak points correspond to moments of sudden load increases, and trough points correspond to moments of sudden load decreases. The time matching algorithm calculates the time interval between the start-up time of each piece of equipment and the peak time of the load fluctuation, as well as the time interval between the stop time of the equipment and the trough time of the load fluctuation, recorded in the equipment start-up and shutdown state matrix. The matching quantification process uses a Gaussian decay function to calculate the matching degree score corresponding to the time interval; the smaller the time interval, the higher the matching degree score, and the larger the time interval, the lower the matching degree score. The constraint weight coefficient is directly proportional to the matching degree score; equipment start-up and shutdown times with high matching degrees are assigned larger constraint weight coefficients, and times with low matching degrees are assigned smaller weight coefficients. The constraint weight coefficients corresponding to each equipment start-up and shutdown state form a weight coefficient vector, which reflects the intensity of the impact of different equipment start-up and shutdown behaviors on the load fluctuation signal.

[0073] The weighted constrained variational objective constructing process weights the high-frequency load fluctuation signals for each time period based on constraint weight coefficients. The weighting process multiplies the constraint weight coefficients as weighting factors with the corresponding load fluctuation signal values ​​at different times. Load fluctuation signals corresponding to equipment start-up and shutdown times are amplified, while signals at other times are reduced. The variational objective function comprises two main components: a data fidelity term and a regularization term. The data fidelity term ensures the consistency between the decomposition result and the original signal, while the regularization term controls the smoothness and sparsity of the modal components. The constrained variational objective constructing algorithm uses the weighted load fluctuation signal as input to the data fidelity term and incorporates equipment state constraints into the regularization term. The equipment-constrained variational optimization objective seeks the optimal variational modal decomposition result by minimizing the weighted sum of the weighted data fidelity error and the constraint regularization term. This objective function ensures that the decomposed modal components are strongly correlated with equipment start-up and shutdown behavior.

[0074] The iterative decomposition strategy employs a combination of frequency domain filtering and time domain constraints for variational mode extraction. Frequency domain filtering decomposes the device-constrained variational optimization objective in the frequency domain by designing filter banks with different center frequencies and bandwidths to decompose the signal into multiple frequency sub-bands. Time domain constraint processing applies device state constraints to each frequency sub-band to ensure that the decomposed modal components exhibit the expected load change characteristics during device start-up and shutdown. The iterative decomposition algorithm alternates between frequency domain filtering and time domain constraint steps. In each iteration, frequency domain filtering updates the frequency characteristics of each mode, while time domain constraints adjust the time characteristics of each mode. Variational mode extraction obtains modal components that satisfy both frequency and time domain constraints by solving the constraint optimization problem. The initial variational mode components are the intermediate output of the iterative decomposition strategy, containing preliminary load change information associated with device start-up and shutdown behavior.

[0075] The modal parameter adjustment process employs gradient descent to optimize the center frequency and bandwidth parameters of the initial variational modal components. The center frequency parameter determines the main frequency components of the modal components, while the bandwidth parameter controls the frequency distribution range of the modal components. An iterative optimization algorithm calculates the gradient direction for parameter adjustment based on the equipment state constraints, with the gradient direction pointing in the direction of parameter change that increases the constraint satisfaction. The parameter adjustment step size is determined using an adaptive strategy: a larger step size is used to accelerate convergence when constraint satisfaction improves significantly, and a smaller step size is used to ensure convergence stability when improvement is slow. The modal parameter adjustment process continues until the constraint satisfaction reaches a preset threshold or the number of iterations reaches its upper limit. The optimized variational modal components are the output of the parameter adjustment process, possessing optimized center frequency and bandwidth parameters, and exhibiting a significantly improved correlation with equipment start-up and shutdown behavior compared to the initial modal components.

[0076] The correlation verification process for equipment start-up and shutdown times calculates the correlation strength between optimized variational modal components and the time nodes of the equipment start-up and shutdown state matrix through correlation analysis. The correlation calculation algorithm uses a cross-correlation function to analyze the correlation between the time series of modal components and the equipment start-up and shutdown state series; the magnitude of the cross-correlation function value reflects the degree of synchronization between the two series in time. The correlation metric process calculates the maximum cross-correlation coefficient between each optimized variational modal component and each equipment start-up and shutdown state; a larger correlation coefficient indicates a higher correlation between the modal component and the corresponding equipment. A filtering process sets a correlation threshold, retaining only modal components with a maximum cross-correlation coefficient exceeding the threshold. A high-correlation modal filtering algorithm further analyzes the modal components that pass the threshold filtering, identifying modal components with significant correlation to specific equipment start-up and shutdown behaviors. The equipment-related variational modal components are the output of the correlation verification process, containing load change characteristics closely related to the start-up and shutdown behaviors of home appliances.

[0077] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0078] Based on the differences in start-up and shutdown frequencies during peak, normal, and valley periods, the equipment start-up and shutdown load correlation characteristics are separated by time period. The equipment start-up and shutdown load correlation data for each time period are extracted to obtain time-segmented start-up and shutdown load characteristic groups.

[0079] Based on the load characteristics of different time periods, a differentiated network structure is constructed to perform time period prediction network configuration processing. A deep network is set for peak periods, a medium network is set for normal periods, and a shallow network is set for valley periods, resulting in a time period network structure configuration.

[0080] The load characteristics of start-up and shutdown in different time periods are input into the corresponding network structure configuration for hyperparameter optimization. The learning rate, number of neurons, and dropout rate of the network in each time period are independently optimized using a successive halving strategy to obtain the time period optimized hyperparameter set.

[0081] Based on the time-period optimized hyperparameter group, the network for each time period is trained to predict start-stop frequency load. The network parameters are learned using historical start-stop frequency data and corresponding load data to obtain the time-period start-stop load prediction network.

[0082] The current start-stop frequency data is input into the time-period start-stop load prediction network for load prediction calculation and processing. The load prediction values ​​for peak period, normal period, and valley period are obtained respectively, and the time-period start-stop load prediction results are obtained.

[0083] Specifically, the time-period feature separation process classifies the equipment start-stop load correlation features based on the differences in start-stop frequency during peak, normal, and valley periods. The difference in start-stop frequency refers to the significant difference in the number of times household appliances start and stop within different time periods. During peak periods, concentrated residential electricity demand leads to frequent equipment start-stops; during normal periods, equipment usage is relatively stable with a moderate number of start-stops; and during valley periods, only basic equipment operates with the fewest start-stops. The separation algorithm extracts data subsets corresponding to the time period from the equipment start-stop load correlation features based on the time period identifier. The peak period data subset includes frequent start-stop information for high-power equipment such as air conditioners and water heaters; the normal period data subset includes start-stop information for intermittent equipment such as washing machines and microwave ovens; and the valley period data subset mainly includes occasional start-stop information for continuously operating equipment such as refrigerators and routers. The time-period start-stop load feature group is the output of the separation process, containing three independent data subsets. Each subset reflects the correlation between equipment start-stop behavior and load changes within the corresponding time period. The differentiated network structure construction process designs corresponding neural network architectures based on the different start-stop frequencies and load complexities of each time period. Deep networks are neural network structures containing multiple hidden layers. With a large number of layers, they can learn complex nonlinear mappings and are suitable for the complex and variable start-stop load patterns during peak hours. Mid-level networks contain a moderate number of hidden layers, balancing learning ability and computational complexity, and are suitable for the relatively stable start-stop load patterns during off-peak hours. Shallow networks contain fewer hidden layers, are simple in structure, and computationally efficient, suitable for the simple start-stop load patterns during off-peak hours. The time-period prediction network configuration involves designing a dedicated network structure for each time period. The peak-hour network uses a deep structure with three hidden layers, the off-peak network uses a mid-level structure with two hidden layers, and the off-peak network uses a shallow structure with one hidden layer. The time-period network structure configuration includes structural parameters such as the number of layers, inter-layer connections, and activation function selection for each time period.

[0084] The hyperparameter optimization process employs a successive halving strategy to independently optimize the key hyperparameters of the network at each time period. The successive halving strategy is an efficient hyperparameter optimization method that finds the optimal configuration by gradually eliminating poorly performing hyperparameter combinations. The algorithm first generates a large number of candidate hyperparameter combinations for each time period, including different learning rate values, number of neurons, and dropout rate settings. The learning rate controls the step size of network parameter updates, the number of neurons determines the network's expressive power, and the dropout rate adjusts the network's regularization strength to prevent overfitting. The optimization process consists of multiple iterations. In each iteration, the algorithm uses training data to train all candidate hyperparameter combinations for a short time. Based on the performance on the validation set, it retains the best-performing half of the candidate combinations and eliminates the poorest half. The successive halving process continues until the number of remaining candidate combinations is small. The algorithm then selects the best-performing combination from the remaining combinations as the optimized hyperparameters for that time period. The time-period optimized hyperparameter set contains the optimized learning rate, number of neurons, and dropout rate parameters for each time period.

[0085] The start-stop frequency load prediction training process uses historical start-stop frequency data and corresponding load data to learn the network parameters for each time period. Historical start-stop frequency data records the number of start-stop cycles for each device during different time periods over a past period, while corresponding load data records the actual power consumption within the same time period. Training data preprocessing uses start-stop frequency data as the network input feature and load data as the network output target. Data is assigned to the corresponding time period network according to the time period identifier for training. The network parameter learning process uses the backpropagation algorithm. The algorithm calculates the parameter gradient based on the error between the predicted load and the actual load, and uses gradient descent to update the network weights and bias parameters. During training, each time period network updates its parameters independently, using optimized hyperparameters for the corresponding time period to control the learning process. The time period start-stop load prediction network is the output of the training process, containing three neural networks trained for peak, normal, and valley periods respectively. The load prediction calculation process inputs the current start-stop frequency data into the corresponding time period network for forward computation. The current start-stop frequency data refers to the statistical count of the number of start-stop cycles for each device within the time period for which load prediction is needed; the data format is consistent with the historical start-stop frequency data used during training. The forecasting process assigns the input data to the corresponding time-period network based on its time-period attributes. Peak-period start / stop frequency data is input into the peak-period network, normal-period data into the normal-period network, and valley-period data into the valley-period network. Each time-period network calculates and outputs the load forecast value for its corresponding time period through forward propagation. The forecast value reflects the expected power consumption for that time period under a given start / stop frequency. The time-period start / stop load forecast results include three independent load forecast values ​​for peak, normal, and valley periods, forming a load forecast sequence covering all time periods throughout the day.

[0086] The household's appliance start-stop load correlation characteristics show that the air conditioner starts and stops more frequently during peak hours, the washing machine starts and stops occasionally during normal hours, and only the refrigerator starts and stops regularly during off-peak hours. The time-period feature separation algorithm divides the correlation features into three subsets based on these start-stop frequency differences. The peak-period subset contains strong correlation information between the air conditioner and load fluctuations; the normal-period subset contains moderate correlation information between the washing machine and load changes; and the off-peak subset contains weak correlation information between the refrigerator and the baseline load. During the differential network structure construction process, a three-layer deep network structure is designed for peak hours due to the complexity of air conditioner start-stop, a two-layer medium-layer network is designed for normal hours due to the regularity of washing machine start-stop, and a single-layer shallow network is designed for off-peak hours due to the simplicity of refrigerator start-stop. In the hyperparameter optimization process, the peak-period network uses a successive halving strategy to select an optimized hyperparameter combination from a large number of candidate combinations. This combination has a learning rate of 1 / 1000, 128 neurons, and a dropout rate of 30%, and performs best on the validation set. During the load prediction training phase based on start-stop frequency, the peak-hour network uses the number of air conditioner start-stop cycles from historical data as input features and the corresponding peak-hour power consumption as the output target for parameter learning. The network learns the mapping relationship between air conditioner start-stop frequency and load changes. When calculating load prediction, the current peak-hour air conditioner start-stop frequency data is input into the trained peak-hour network. The network calculates the corresponding load prediction value based on the learned mapping relationship. This prediction value reflects the impact of air conditioner start-stop behavior on peak-hour power consumption. This time-period-based network design and training strategy solves the problem of poor model adaptability in existing technologies. By designing specialized network structures and optimization strategies for different time periods, it avoids the technical deficiency of a single model being unable to simultaneously handle the differences in load characteristics across different time periods. Furthermore, independent hyperparameter optimization ensures that the network reaches its optimal performance state for each time period.

[0087] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0088] Load continuity constraint calculations are performed on the switching points of peak, normal, and valley periods in the load forecast results for time-based start and stop, and load continuity constraint conditions between adjacent time periods are established to obtain the time-based switching continuity constraint parameters.

[0089] Based on the equipment start-stop state matrix, the state transfer matrix is ​​constructed to analyze the equipment state transfer relationship between different time periods. The state influence weight of the equipment between adjacent time periods is calculated to obtain the equipment state transfer weight matrix.

[0090] Based on the continuity constraint parameters of time period switching and the weight matrix of equipment status transfer, the start-stop load prediction results of time periods are weighted and fused to calculate the weight, and time-related fusion weights are assigned to the prediction results of each time period to obtain the time period fusion weight coefficient.

[0091] The load forecast results for each time period are weighted and summed according to the time period fusion weight coefficient. The combined load forecast sequence is obtained by combining the weight coefficient of the impact of equipment start-up and shutdown frequency on the load.

[0092] The equipment status consistency test is performed on the fused load prediction sequence to verify the logical consistency between the fused result and the actual equipment start-up and shutdown status. Iterative adjustments and optimizations are then performed to obtain the time-of-use power load modeling result.

[0093] Specifically, the load continuity constraint calculation process establishes constraints by analyzing the load jump amplitude at the switching points of the start-stop load forecast results during peak, normal, and valley periods. The time switching point refers to the time boundary between adjacent time periods, such as the moment when the peak period ends and the normal period begins. Load continuity means that the load value change before and after the time switching point should transition smoothly, avoiding unreasonable sudden jumps. The constraint calculation algorithm first identifies the load value of each time period's forecast results at the switching point and calculates the difference between the load forecast value before and after the switching point. The continuity constraint requires that the absolute value of this difference does not exceed a specific proportion of the average load of that time period, typically set to 15% of the average load. The time period switching continuity constraint parameters include the load difference limit threshold and constraint strength coefficient for each switching point; these parameters specify the continuity requirements that need to be met during the fusion process. The state transfer matrix construction process analyzes the transfer law of equipment operating states between time periods based on the equipment start-stop state matrix. The equipment state transfer relationship refers to the degree of influence of the equipment's operating state in the previous time period on the operating state in the next time period, such as the impact of frequent start-stop of air conditioners during peak periods on the start-stop probability during normal periods. An algorithm is constructed to statistically analyze the frequency of device state transitions between adjacent time periods in historical data, and to calculate the probability distribution of various states from a certain state in time period A to a certain state in time period B. The state influence weight is calculated using conditional probability, reflecting the strength of the influence of the device state in the previous time period on the device state and load changes in the next time period. The device state transfer weight matrix is ​​a three-dimensional data structure: the first dimension represents the device type, the second dimension represents the source time period, and the third dimension represents the target time period. The matrix element values ​​represent the magnitude of the state transfer weight of the corresponding device between the corresponding time periods.

[0094] The weighted fusion weight calculation process assigns fusion weights to the prediction results of each time period based on the time-switching continuity constraint parameters and the equipment status transfer weight matrix. Time-related fusion weights determine the contribution of the prediction result of a given time period to the predicted value at that time period's center time, based on the distance between the predicted time and the center time of that time period. The weight calculation algorithm uses a Gaussian decay function, assigning higher weights to times closer to the center time period and lower weights to times farther away. The equipment status transfer weight matrix adjusts the weight distribution between adjacent time periods; when the equipment status transfer weight is large, the influence of the previous time period on the subsequent time period is enhanced, and the weight allocation is tilted towards the previous time period. The continuity constraint parameters affect the weight calculation through a penalty function; when the prediction result violates the continuity constraint, the weight of the corresponding time period is reduced. The time-switching fusion weight coefficient is the output of the weight calculation process, containing the fusion weight allocation scheme for each time period at different times. The weighted summation integration process linearly combines the time-switching start-stop load prediction results according to the time-switching fusion weight coefficient. Weighted summation means multiplying the predicted value of each time period by its corresponding fusion weight and then summing the results to obtain the comprehensive prediction value at each time period. The ensemble algorithm iterates through each moment of the forecast time series, extracting the predicted values ​​and corresponding fusion weights for each time period, and performs a weighted summation calculation. The weighting coefficient for the impact of equipment start-up and shutdown frequency on load is used as an additional adjustment factor in the comprehensive calculation; this coefficient reflects the importance of equipment start-up and shutdown behavior to load changes. The comprehensive calculation process multiplies the time-period fusion result with the equipment start-up and shutdown impact weights to form a fusion forecast value that considers both time-period characteristics and equipment behavior. The fusion load forecast sequence is the output of the ensemble processing, containing the load forecast time series after time-period fusion and equipment impact adjustment.

[0095] The equipment status consistency verification process verifies the rationality of the fusion result by comparing the degree of matching between the fused load forecast sequence and the actual equipment start-up and shutdown status. Logical consistency means that the predicted load changes should maintain a logical correspondence with the actual equipment start-up and shutdown behavior; the equipment start-up time corresponds to a load increase, and the equipment shutdown time corresponds to a load decrease. The verification algorithm extracts the start-up and shutdown times recorded in the equipment start-up and shutdown status matrix and finds the corresponding load change direction in the fused load forecast sequence. The consistency verification calculates the matching degree between the equipment start-up and shutdown direction and the load change direction; a high matching degree indicates that the fusion result maintains good logical consistency with the actual equipment behavior. The iterative adjustment and optimization algorithm adjusts the fusion weight coefficients based on the consistency verification results. When a moment with poor consistency is found, the algorithm reduces the fusion weight of the dominant period at that moment and increases the weight of other periods. The time-of-use load modeling result is the output of the consistency verification and iterative optimization, including the verified and adjusted load forecast sequence and the corresponding confidence assessment.

[0096] The above describes the modeling method for the electricity load model based on time-of-use consumption in the embodiments of this application. The following describes the modeling system for the electricity load model based on time-of-use consumption in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the electricity load modeling system based on time-of-use electricity consumption in this application includes:

[0097] The data acquisition module is used to collect time-of-use electricity consumption data from residential users through smart meters to obtain time-of-use electricity consumption sequences. At the same time, the device monitoring module identifies the start-stop frequency of household appliances to obtain a device start-stop status matrix.

[0098] The decomposition module is used to perform differential decomposition processing on the data of each time period based on the time-of-use electricity sequence and the time-of-use electricity empirical mode decomposition algorithm to obtain the basic load characteristics of the peak-valley and normal periods.

[0099] The identification module is used to identify and process the start-stop fluctuation signal by using the time-division variational mode decomposition algorithm to identify the basic load characteristics, and to obtain the equipment start-stop load correlation characteristics by combining the equipment start-stop state matrix.

[0100] The prediction module is used to perform load prediction processing on the start-up and stop frequencies of each time period based on the start-up and stop load correlation characteristics of the equipment, and to obtain the start-up and stop load prediction results for each time period.

[0101] The integration module is used to integrate the start-stop load prediction results of the time period through the time-sharing load status fusion algorithm, establish the start-stop frequency influence weight mechanism, and obtain the time-sharing power load modeling results.

[0102] above Figure 2 The electricity load modeling system based on time-of-use electricity consumption in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The electricity load modeling device based on time-of-use electricity consumption in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0103] Reference Figure 3 This invention also provides a time-of-use (TOU) power load modeling device, which can be a server, and its internal structure can be as follows: Figure 3As shown, the time-of-use (TOU) electricity load modeling device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory of the TOU electricity load modeling device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the TOU electricity load modeling device stores the data corresponding to this embodiment. The network interface of the TOU electricity load modeling device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0104] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the modeling device for the time-of-use electricity load model applied thereto.

[0105] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for modeling the electricity load model based on time-of-use electricity.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a time-of-use electricity load modeling device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] 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.

Claims

1. A method for modeling electricity load based on time-of-use electricity consumption, characterized in that, The method includes: By collecting time-of-use electricity data from residential users through smart meters, a time-of-use electricity sequence is obtained. At the same time, the start-stop frequency of household appliances is identified through the device monitoring module, resulting in a device start-stop status matrix. Based on the time-of-use electricity sequence, the time-of-use electricity empirical mode decomposition algorithm is used to perform differential decomposition processing on the data of each time period to obtain the basic load characteristics of peak-valley and normal periods. The basic load characteristics are processed by time-division variational mode decomposition algorithm to identify start-stop fluctuation signals, and combined with the equipment start-stop state matrix, the equipment start-stop load correlation characteristics are obtained. Based on the equipment start-up and shutdown load correlation characteristics, a time-sharing hyperparameter optimization algorithm is used to perform load prediction processing on the start-up and shutdown frequencies of each time period to obtain the time-sharing start-up and shutdown load prediction results. This includes: performing time-sharing feature separation processing on the equipment start-up and shutdown load correlation characteristics based on the differences in start-up and shutdown frequencies during peak, normal, and valley periods; extracting equipment start-up and shutdown load correlation data for each time period to obtain time-sharing start-up and shutdown load feature groups; constructing a differentiated network structure based on the time-sharing start-up and shutdown load feature groups for time-sharing prediction network configuration processing, setting a deep network for peak periods, a mid-level network for normal periods, and a shallow network for valley periods to obtain the time-sharing network structure configuration; and then configuring the time-sharing start-up and shutdown load prediction network. The load outage feature groups are input into the corresponding time-segment network structure configurations for hyperparameter optimization. A successive halving strategy is used to independently optimize the learning rate, number of neurons, and dropout rate of the network for each time segment, resulting in a time-segment optimized hyperparameter group. Based on the time-segment optimized hyperparameter group, the network for each time segment is trained for start-stop frequency load prediction. Historical start-stop frequency data and corresponding load data are used to learn the network parameters, resulting in a time-segment start-stop load prediction network. The current start-stop frequency data is input into the time-segment start-stop load prediction network for load prediction calculation, obtaining load prediction values ​​for peak, normal, and valley periods, respectively, to obtain the time-segment start-stop load prediction results. The start-stop load prediction results for the time period are integrated and processed by the time-sharing load status fusion algorithm, and a start-stop frequency influence weighting mechanism is established to obtain the time-sharing electricity load modeling results.

2. The method for modeling electricity load based on time-of-use electricity consumption according to claim 1, characterized in that, The method involves collecting time-of-use electricity consumption data from residential users using smart meters to obtain a time-of-use electricity consumption sequence. Simultaneously, a device monitoring module identifies the start-stop frequency of household appliances to obtain a device start-stop status matrix, including: The 24-hour time period is divided into peak, normal and off-peak periods to obtain three time-of-use electricity price intervals. Based on the three time-of-use electricity price periods, the electricity data collected by the smart meter is sampled at 15-minute intervals to obtain the time-of-use electricity data points for each period. The time-of-use electricity data points for each time period are arranged and combined according to the time sequence to obtain a time-of-use electricity sequence that includes peak, valley, and flat load changes; Based on the power threshold of household appliances, air conditioners, water heaters, and washing machines are screened for high-power equipment to obtain a list of target monitoring equipment. Based on the target monitoring equipment list, the start-stop status of each equipment is monitored by current sensors, and the number of times the equipment is started and stopped is recorded to obtain the equipment start-stop status matrix.

3. The method for modeling electricity load based on time-of-use electricity consumption according to claim 1, characterized in that, The step involves using the time-of-use electricity consumption sequence and employing a time-of-use electricity consumption empirical mode decomposition algorithm to perform differential decomposition processing on the data for each time period, thereby obtaining the basic load characteristics for peak-valley and normal periods, including: Based on the load fluctuation variance of peak, normal, and valley periods, the time-of-use electricity sequence is subjected to adaptive noise amplitude calculation to obtain differentiated noise parameters for each time period; Based on the differentiated noise parameters for each time period, paired white noise is added to the corresponding time period's electricity data for noise enhancement processing, resulting in a noisy time period electricity sequence. The noisy time-period electricity sequence is processed by empirical mode decomposition according to the number of high-frequency modes during peak periods, the number of mid-frequency modes during normal periods, and the number of low-frequency modes during valley periods to obtain the inherent mode components of each time period. The inherent modal components of each time period are integrated and averaged to eliminate the influence of noise and retain the true load change trend, thus obtaining the time period purification modal components. Based on the frequency characteristics of the purification mode components during the specified time period, the basic load mode screening process is performed to extract the mode components that reflect the basic electricity consumption patterns during each time period, thereby obtaining the basic load characteristics during peak, valley, and normal periods.

4. The method for modeling electricity load based on time-of-use electricity consumption according to claim 1, characterized in that, The process of identifying start-stop fluctuation signals by applying a time-division variational mode decomposition algorithm to the basic load characteristics, and combining this with the equipment start-stop state matrix, yields the equipment start-stop load correlation characteristics, including: High-frequency modal components are extracted from the basic load characteristics of the peak-valley and normal periods, and start-stop fluctuation signals are filtered to obtain high-frequency load fluctuation signals for each time period. Based on the differences in load characteristics during peak, normal, and valley periods, differentiated penalty parameters are set for the high-frequency load fluctuation signals of each time period to obtain time-period differentiated decomposition parameters. Based on the equipment start-stop state matrix, equipment state constraints are constructed and a constraint mechanism is established. The equipment start-stop state is associated with the load fluctuation signal to obtain the equipment state constraints. The high-frequency load fluctuation signals of each time period and the equipment state constraints are input into the variational mode decomposition algorithm for constraint variational solution processing. The optimal variational mode is solved through iterative optimization to obtain the equipment-related variational mode components. The device-associated variational mode components are subjected to start-stop behavior matching and verification processing to identify the load change modes corresponding to the start-stop times of the home appliances, thereby obtaining the device start-stop load association features.

5. The method for modeling electricity load based on time-of-use electricity consumption according to claim 4, characterized in that, The process involves inputting the high-frequency load fluctuation signals of each time period and the equipment state constraints into a variational mode decomposition algorithm for constraint variational solution processing. Through iterative optimization, the optimal variational mode is solved to obtain the equipment-related variational mode components, including: Based on the equipment state constraints, the constraint weight coefficients of the high-frequency load fluctuation signals in each time period are calculated and processed. The equipment start-up and shutdown times are matched and quantified with the peak load fluctuation times to obtain the constraint weight coefficients corresponding to each equipment start-up and shutdown state. Based on the constraint weight coefficients, the high-frequency load fluctuation signals of each time period are processed by weighted constraint variational objective construction, and the load fluctuation signals corresponding to the equipment start-up and shutdown times are given higher weights to obtain the equipment constraint variational optimization objective. The device-constrained variational optimization objective is subjected to an iterative decomposition strategy for variational mode extraction. Each variational mode is separated by a combination of frequency domain filtering and time domain constraints to obtain the initial variational mode components. The initial variational mode components are processed by an iterative optimization algorithm to adjust the mode parameters. The center frequency and bandwidth parameters of each mode are continuously adjusted according to the equipment state constraints to obtain the optimized variational mode components. The optimized variational mode components are subjected to equipment start-up and shutdown time correlation verification processing. The correlation between each mode and the time nodes of the equipment start-up and shutdown state matrix is ​​calculated and high correlation modes are selected to obtain the equipment-related variational mode components.

6. The method for modeling electricity load based on time-of-use electricity consumption according to claim 1, characterized in that, The process involves integrating the start-stop load prediction results for the specified time period using a time-of-use load state fusion algorithm, establishing a start-stop frequency influence weighting mechanism, and obtaining the time-of-use electricity load modeling results, including: The load continuity constraint calculation is performed on the switching points of peak, normal, and valley periods in the load start-stop prediction results of the time period, and load continuity constraint conditions between adjacent time periods are established to obtain the time period switching continuity constraint parameters. Based on the device start / stop state matrix, the state transfer matrix is ​​constructed for the device state transfer relationship between each time period. The state influence weight of the device between adjacent time periods is calculated to obtain the device state transfer weight matrix. Based on the time period switching continuity constraint parameters and the equipment state transfer weight matrix, the time period start-stop load prediction results are weighted and fused to calculate the weight, and time-related fusion weights are assigned to the prediction results of each time period to obtain the time period fusion weight coefficient. The start-stop load prediction results for the specified time period are weighted and summed according to the time period fusion weight coefficient. Combined with the weight coefficient of the impact of equipment start-stop frequency on load, a comprehensive calculation is performed to obtain the fusion load prediction sequence. The fused load prediction sequence is subjected to equipment status consistency verification to verify the logical consistency between the fused result and the actual equipment start-stop status. Iterative adjustments and optimizations are then performed to obtain the time-of-use power load modeling result.

7. A modeling system for electricity load based on time-of-use electricity consumption, characterized in that, The method for modeling an electricity load model based on time-of-use electricity consumption as described in any one of claims 1-6, wherein the electricity load modeling system based on time-of-use electricity consumption comprises: The data acquisition module is used to collect time-of-use electricity consumption data from residential users through smart meters to obtain time-of-use electricity consumption sequences. At the same time, the device monitoring module identifies the start-stop frequency of household appliances to obtain a device start-stop status matrix. The decomposition module is used to perform differential decomposition processing on the data of each time period based on the time-sharing power consumption sequence using the time-sharing power consumption empirical mode decomposition algorithm, so as to obtain the basic load characteristics of the peak-valley and normal periods. The identification module is used to process the start-stop fluctuation signal identification of the basic load characteristics through the time-division variational mode decomposition algorithm, and combine it with the equipment start-stop state matrix to obtain the equipment start-stop load correlation characteristics. The prediction module is used to perform load prediction processing on the start-stop frequency of each time period according to the equipment start-stop load correlation characteristics, using a time-sharing hyperparameter optimization algorithm to obtain the time-sharing start-stop load prediction results. This includes: performing time-sharing feature separation processing on the equipment start-stop load correlation characteristics based on the differences in start-stop frequencies during peak, normal, and valley periods; extracting equipment start-stop load correlation data for each time period to obtain time-sharing start-stop load feature groups; constructing a differentiated network structure based on the time-sharing start-stop load feature groups for time-sharing prediction network configuration processing, setting a deep network for peak periods, a medium-layer network for normal periods, and a shallow network for valley periods to obtain a time-sharing network structure configuration; and then configuring the time-sharing network structure. The start-stop load feature groups for each time period are input into the corresponding time-segment network structure configuration for hyperparameter optimization. A successive halving strategy is used to independently optimize the learning rate, number of neurons, and dropout rate of the network for each time period, resulting in a time-segment optimized hyperparameter group. Based on the time-segment optimized hyperparameter group, the network for each time period is trained for start-stop frequency load prediction. Historical start-stop frequency data and corresponding load data are used to learn the network parameters, resulting in a time-segment start-stop load prediction network. The current start-stop frequency data is input into the time-segment start-stop load prediction network for load prediction calculation, obtaining the load prediction values ​​for peak, normal, and valley periods, respectively, to obtain the time-segment start-stop load prediction results. The integration module is used to integrate the start-stop load prediction results of the time period through the time-sharing load status fusion algorithm, establish the start-stop frequency influence weight mechanism, and obtain the time-sharing power load modeling results.

8. A modeling device for electricity load based on time-of-use electricity consumption, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the power load modeling method based on time-of-use electricity as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the power load modeling method based on time-of-use electricity as described in any one of claims 1 to 6.

Images