Electricity load prediction and intelligent suggestion method and system, electric energy meter and storage terminal

Abstract

The application relates to the technical field of data processing, and discloses a power consumption load prediction and intelligent suggestion method and system, an electric energy meter and a storage terminal. The power consumption load prediction and intelligent suggestion method comprises the following steps: obtaining historical power consumption data, user behavior data, environment data, time data and electricity price information; performing pretreatment operation on the historical power consumption data, the user behavior data, the environment data and the time data to obtain a power consumption load time sequence; determining a load prediction curve according to the power consumption load time sequence and a first prediction model which is a hybrid model composed of a variational mode decomposition model, a long sequence time sequence prediction model and a long short-term memory network model; and generating a power consumption suggestion according to the load prediction curve, the user behavior data and the electricity price information. The application realizes accurate prediction of user power consumption load, and generates a power consumption suggestion based on the predicted user power consumption load.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for predicting and intelligently recommending electricity load, an electricity meter, and a storage terminal. Background Technology

[0002] Traditional electricity meters are metering devices used to measure and record the total amount of electricity consumed by users. Their function is singular, and users cannot access detailed information about their electricity usage. Smart meters, in addition to metering, also have functions such as data storage, load identification, and event logging. They can be viewed in real time via mobile apps and other terminals, achieving a high degree of transparency in electricity usage and greater interactivity with users. With the deepening of energy transformation and upgrading, the power system is accelerating its evolution towards digitalization, clean energy, and intelligence. As a core tool driving this transformation, the digitalization level of smart meters plays a crucial role in improving the overall informatization level of the power system.

[0003] The State Grid Corporation of China is actively responding to national strategies and has proposed using artificial intelligence (AI) as a key tool to promote the construction of a new power system and achieve high-quality development. Currently, AI technology is not widely applied to electricity meters to achieve intelligent processing of electricity consumption data. Applying AI technology to electricity meters, based on load forecasting and user behavior identification, can provide intelligent electricity consumption suggestions. By deeply mining the value of metering data, it can optimize the user's electricity experience, promote energy conservation and emission reduction, and drive the intelligent transformation of the energy system.

[0004] However, the relevant technologies lack accurate methods for predicting the short-term load of electricity meter users, and smart meters have not yet introduced the function of providing users with smart electricity consumption suggestions based on load forecasting. Summary of the Invention

[0005] The present invention aims to solve the technical problems existing in the prior art or related technologies, such as the lack of accurate prediction of the short-term load of electricity meter users and the inability to provide users with smart electricity consumption suggestions.

[0006] Therefore, the first aspect of the present invention proposes a method for predicting and intelligently recommending electricity load.

[0007] A second aspect of the present invention proposes an electricity load prediction and intelligent suggestion system.

[0008] A third aspect of the present invention provides an electricity meter.

[0009] A fourth aspect of the present invention provides a storage terminal.

[0010] In view of this, according to a first aspect of the present invention, a method for electricity load forecasting and intelligent suggestion is proposed. The method includes: acquiring historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information; performing preprocessing operations on the historical electricity consumption data, user behavior data, environmental data, and time data to obtain an electricity load time series sequence; determining a load forecast curve based on the electricity load time series sequence and a pre-set first forecast model, wherein the first forecast model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series forecast model, and a long short-term memory network model; and generating electricity consumption suggestions based on the load forecast curve, user behavior data, and electricity price information.

[0011] The electricity load forecasting and intelligent suggestion method provided by this invention includes: First, acquiring historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information. Then, preprocessing the historical electricity consumption data, user behavior data, environmental data, and time data to obtain an electricity load time-series sequence. The electricity load time-series sequence refers to a continuous sequence of electricity consumption values ​​arranged sequentially based on time. The preprocessing operation involves integrating all scattered data at a unified time granularity, removing interfering data collected temporarily, and standardizing the data value range. This transforms the chaotic and irregular discrete electricity consumption data into time-series structured data that is continuously arranged in chronological order, has a unified format, and can be directly input into model calculations. Subsequently, after obtaining the electricity load time series, a load prediction curve is determined based on the electricity load time series and a pre-set first prediction model. The first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model, and a long short-term memory network model. In other words, the first prediction model is an integrated prediction model that combines signal decomposition algorithms, long-sequence prediction models, and short-sequence prediction networks, employing dual-path synchronous computation. It can accurately predict both long-term electricity consumption trends and short-term fluctuations. After obtaining the first prediction model, the electricity load time series is input into it. Relying on the built-in dual-branch independent computation logic of the first prediction model, different variation patterns within the data are analyzed separately. The output results of the two computations are integrated to draw a continuous, complete, and accurate long-term electricity load variation curve that closely reflects actual electricity consumption fluctuations. Finally, intuitive electricity fluctuation characteristics within the electricity load variation curve are extracted. Combined with user behavior data such as daily start-up and shutdown of electrical equipment and fixed-time electricity consumption, and in accordance with electricity price information, targeted and actionable electricity consumption recommendations are formulated. This invention comprehensively covers factors influencing electricity consumption by relying on multi-dimensional basic data, including historical electricity consumption data, user behavior data, environmental data, and time data. Preprocessing eliminates computational errors caused by data deficiencies, and a combined prediction model (the first prediction model) achieves accurate predictions that balance long-term electricity consumption trends with short-term fluctuations. Finally, it combines actual usage scenarios to provide rational suggestions. This achieves automated and accurate prediction of user electricity load, and generates personalized electricity guidance by combining users' actual electricity consumption habits with electricity pricing policies. It significantly reduces the workload of manual statistical analysis of electricity consumption data and improves the rationality of electricity planning.

[0012] In some technical solutions, optionally, the step of determining the load forecast curve based on the electricity load time series and a pre-set first forecast model includes: determining a first mode component and a second mode component based on the electricity load time series and a variational mode decomposition model, wherein the frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than a preset frequency; determining first load change forecast data based on the first mode component and a long short-term memory network model; determining second load change forecast data based on the second mode component and a long-sequence time series forecast model; and determining the load forecast curve based on the first load change forecast data and the second load change forecast data.

[0013] In this technical solution, the steps of determining the load prediction curve based on the electricity load time series and the first prediction model include: firstly, inputting the electricity load time series into the variational mode decomposition model, which adaptively decomposes the original non-stationary, highly complex electricity load time series into a series of intrinsic mode components with different center frequencies and finite bandwidths. Then, these intrinsic mode components are divided into two categories according to frequency: the first category is the first mode component, and the second category is the second mode component. The frequency of the first mode component is greater than or equal to the preset frequency, and the frequency of the second mode component is less than the preset frequency. That is to say, the first mode component is a short-period, high-frequency, and highly volatile intrinsic mode component, and the second mode component is a long-period, low-frequency, and clearly trending intrinsic mode component. After obtaining the first and second modal components, the high-frequency first modal component is input separately into the Long Short-Term Memory (LSTM) network model. Relying on the gating structure within the LTM model, it focuses on learning short-term patterns such as sudden fluctuations in electricity consumption and temporary increases / decreases in electricity demand, outputting predictive data that accurately reflects short-term electricity consumption fluctuations—the first load change prediction data. Subsequently, the low-frequency second modal component is input into the long-sequence time-series prediction model. Relying on the sparse attention mechanism built into the long-sequence time-series prediction model, it mines stable electricity consumption cycle patterns and seasonal electricity consumption variation patterns over long periods, outputting predictive data that reflects the overall long-term electricity consumption trend—the second load change prediction data. Finally, the first and second load change prediction data are integrated to obtain the final load prediction curve. This invention utilizes a variational mode decomposition model to accurately decompose signal features, thereby separating different electricity consumption change patterns. Then, it matches the corresponding prediction model with the strongest adaptability to learn specific patterns, avoiding the shortcomings of a single model that cannot simultaneously take into account both long and short time series patterns. Finally, it integrates the two independent prediction results, taking into account both the global trend and local fluctuations, making the final prediction results more consistent with real electricity consumption scenarios.

[0014] In some technical solutions, optionally, the step of determining the load forecast curve based on the first load change forecast data and the second load change forecast data includes: weighting and unifying the first load change forecast data and the second load change forecast data to obtain the third load change forecast data; and performing mapping calculations and numerical restoration on the third load change forecast data to obtain the load forecast curve.

[0015] In this technical solution, the steps for determining the load forecast curve based on the first load change forecast data and the second load change forecast data include: First, weighting and unifying the first load change forecast data and the second load change forecast data to obtain the third load change forecast data. That is, firstly, based on the proportion of the impact of long-term electricity consumption trends and short-term electricity fluctuations on the overall load, a differentiated fusion weight coefficient is set. Then, the first load change forecast data and the second load change forecast data are weighted and calculated according to the fusion weight coefficient. At the same time, the data dimension, data value format and time sampling interval of the first load change forecast data and the second load change forecast data are unified to eliminate the problems of inconsistent output data formats and misaligned time nodes of different models. After obtaining the third load change prediction data, the original electricity consumption data underwent normalization and data reduction in the previous preprocessing operations. Therefore, the fused third load change prediction data is not the actual value of the electricity load. At this point, a preset reverse mapping calculation formula is used to restore the standardized calculation data to the actual electricity consumption and power consumption values ​​that can be directly used in the real scenario. Then, based on the restored real continuous values, a visualized load prediction curve is generated according to the time sequence arrangement rules, and key peak and valley nodes of electricity consumption within the curve are marked. This invention distinguishes the degree of influence of different electricity consumption patterns through weighted fusion, ensuring that the integration results are clear in their primary and secondary aspects. It eliminates data interaction barriers by relying on dimensional unification, and then uses reverse numerical mapping rules to offset the numerical scaling effect caused by the previous data preprocessing operations, accurately restoring the actual electricity load values ​​and ensuring that the visualized curve data has practical application value.

[0016] In some technical solutions, optionally, before determining the load forecast curve based on the electricity load time series and a pre-set first forecast model, the electricity load forecasting and intelligent suggestion method includes: determining the parameter combinations that need to be optimized among the variational mode decomposition model, the long-sequence time series forecast model, and the long short-term memory network model; optimizing the parameter combinations using a sparrow search algorithm to obtain optimized parameter combinations; optimizing the variational mode decomposition model, the long-sequence time series forecast model, and the long short-term memory network model respectively based on the optimized parameter combinations; and constructing a first forecast model based on the optimized variational mode decomposition model, the optimized long-sequence time series forecast model, and the optimized long short-term memory network model.

[0017] In this technical solution, before determining the load forecast curve based on the electricity load time series and a pre-set first prediction model, the electricity load forecasting and intelligent suggestion method further includes: First, pre-building the basic blank network architectures of the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model, respectively. Then, determining the parameter combinations that need optimization among the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model, where each parameter combination includes multiple parameters belonging to these three models. Subsequently, based on the parameter combinations, a sparrow search algorithm is used for continuous iterative updates, thereby globally selecting parameter combinations among the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model that are suitable for the current electricity consumption data characteristics. The sparrow search algorithm is an intelligent optimization algorithm that simulates the foraging and early warning escape behavior of sparrows in nature. After obtaining the optimized parameter combinations, the parameters from these combinations are substituted into the basic blank network architectures of the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model, respectively, to optimize these models. Finally, the optimized variational mode decomposition model, the optimized long-sequence time-series prediction model, and the optimized long short-term memory network model are combined to obtain the first prediction model. This invention first establishes a standardized and general basic model architecture to ensure the overall framework's versatility. Then, relying on the powerful global search capability of the sparrow search algorithm, it quickly identifies a suitable configuration scheme within a massive range of parameter values. Finally, the suitable configuration scheme is combined with the basic architecture, allowing the final model to perfectly adapt to the characteristics of the electricity consumption time-series data to be processed, thus ensuring model performance from both architectural and parameter perspectives.

[0018] In some technical solutions, the parameter combinations may optionally include: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time-series prediction model, and the number of attention heads in the long sequence time-series prediction model.

[0019] In this technical solution, parameter combinations may include: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time series prediction model, and the number of attention heads in the long sequence time series prediction model. By classifying the core parameters according to three major categories—variational mode decomposition model, long short-term memory network model, and long sequence time series prediction model—the key configuration items affecting prediction accuracy and computation speed during the operation of the first prediction model are comprehensively covered. This allows the Sparrow Search algorithm to accurately focus on the core parameters for optimization, ensuring that the final parameter combination can comprehensively optimize the overall performance of the model.

[0020] In some technical solutions, optionally, the step of generating electricity consumption recommendations based on load forecast curves, user behavior data, and electricity price information includes: determining peak and off-peak electricity consumption periods based on load forecast curves; determining user electricity consumption habits based on peak and off-peak electricity consumption periods and user behavior data; and generating electricity consumption recommendations based on user electricity consumption habits and electricity price information.

[0021] In this technical solution, the steps for generating electricity consumption recommendations based on load forecast curves, user behavior data, and electricity price information include: First, determining peak and off-peak electricity consumption periods based on the load forecast curve. That is, relying on the long-term load forecast curve, a unified load value threshold is set. Ranges with values ​​above the threshold are defined as peak consumption periods, and ranges with values ​​below the threshold are defined as off-peak periods. Simultaneously, the duration and peak load of peak and off-peak periods are statistically analyzed, and the distribution patterns of peak and off-peak electricity consumption periods on weekdays and weekends can be distinguished. Then, combining peak and off-peak periods with user behavior data, users' daily electricity consumption habits are determined. Subsequently, based on these habits, the defined peak and off-peak periods are matched, and users' regular electricity consumption habits, such as fixed high-power appliance usage periods, basic household electricity consumption periods, and idle / power-off periods, are summarized. Finally, referring to the electricity price information, i.e., the grid's peak-valley time-of-use pricing standard, and considering users' electricity consumption habits, targeted electricity consumption recommendations are given without significantly altering users' normal life and production electricity consumption order. This invention uses accurate load forecast curves as the basis for time patterns, real user electricity consumption data as the basis for usage habits, and official electricity pricing rules as the basis for economic regulation. The combination of these three data sources leads to an optimized electricity consumption plan that not only conforms to the actual electricity consumption situation but also aligns with electricity pricing policy guidelines, thus balancing practicality and economy.

[0022] In some technical solutions, the preprocessing operations may optionally include one or a combination of the following: missing value interpolation completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment and processing of multi-source data.

[0023] In this technical solution, preprocessing operations may include one or a combination of the following: missing value interpolation and completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment of multi-source data. Specifically, missing value interpolation and completion refers to filling in data gaps by extrapolating from adjacent normal data. Abnormal noise data removal refers to identifying and deleting erroneously collected data that deviates from the normal range of electricity consumption fluctuations. Multi-dimensional data normalization refers to converting data of different magnitudes, such as voltage, electricity consumption, temperature, and humidity, to the same numerical range. Dimensionality reduction of invalid and redundant features refers to deleting duplicate collected data that has no practical use in analyzing electricity consumption patterns, thus reducing the amount of data processing. Time-series alignment of multi-source data refers to unifying the collection time nodes of electricity consumption data, environmental data, and time data to achieve complete time-series matching of multiple data types.

[0024] According to a second aspect of the present invention, an electricity load forecasting and intelligent suggestion system is proposed, wherein the electricity load forecasting and intelligent suggestion system includes: a first acquisition module, which is used to acquire historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information; a preprocessing module, which is used to preprocess the historical electricity consumption data, user behavior data, environmental data, and time data to obtain an electricity load time series sequence; a first determination module, which is used to determine a load forecast curve based on the electricity load time series sequence and a pre-set first prediction model, wherein the first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model, and a long short-term memory network model; and a generation module, which is used to generate electricity suggestions based on the load forecast curve, user behavior data, and electricity price information.

[0025] The electricity load forecasting and intelligent suggestion system provided by this invention mainly includes: a first acquisition module, a preprocessing module, a first determination module, and a generation module. The first acquisition module first acquires historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information. Subsequently, the preprocessing module performs preprocessing operations on the historical electricity consumption data, user behavior data, environmental data, and time data to obtain an electricity load time-series sequence. This electricity load time-series sequence refers to a continuous sequence of electricity consumption values ​​arranged sequentially based on time. The preprocessing operation involves integrating all scattered data at a unified time granularity, removing interfering data collected temporarily, and standardizing the data value range. This transforms the chaotic and irregular discrete electricity consumption data into time-series structured data that is continuously arranged in chronological order, has a unified format, and can be directly input into model calculations. Subsequently, after obtaining the electricity load time series, the first determination module determines the load prediction curve based on the electricity load time series and the pre-set first prediction model. The first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model, and a long short-term memory network model. In other words, the first prediction model is an integrated prediction model that combines signal decomposition algorithms, long-sequence prediction models, and short-sequence prediction networks, employing dual-path synchronous computation. It can accurately predict both long-term electricity consumption trends and short-term fluctuations. After obtaining the first prediction model, the electricity load time series is input into it. Relying on the built-in dual-branch independent computation logic of the first prediction model, different variation patterns within the data are analyzed separately. The output results of the two computations are integrated to draw a continuous, complete, and accurate long-term electricity load change curve that closely reflects actual electricity fluctuation patterns. Finally, the generation module extracts intuitive electricity fluctuation characteristics from the electricity load change curve and, combined with user behavior data such as daily start-up and shutdown of electrical equipment and fixed-time electricity consumption, and in accordance with electricity price information, formulates targeted and actionable electricity consumption recommendations. This invention comprehensively covers factors influencing electricity consumption by relying on multi-dimensional basic data, including historical electricity consumption data, user behavior data, environmental data, and time data. Preprocessing eliminates computational errors caused by data deficiencies, and a combined prediction model (the first prediction model) achieves accurate predictions that balance long-term electricity consumption trends with short-term fluctuations. Finally, it combines actual usage scenarios to provide rational suggestions. This achieves automated and accurate prediction of user electricity load, and generates personalized electricity guidance by combining users' actual electricity consumption habits with electricity pricing policies. It significantly reduces the workload of manual statistical analysis of electricity consumption data and improves the rationality of electricity planning.

[0026] According to a third aspect of the present invention, an electricity meter is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the electricity load prediction and intelligent suggestion method as described above.

[0027] The electricity meter provided by the present invention includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the steps of the above-mentioned electricity load prediction and intelligent suggestion method, and can achieve the technical effects of any of the above technical solutions, which will not be elaborated here.

[0028] According to a fourth aspect of the invention, a storage terminal is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the power load prediction and intelligent suggestion method as described above.

[0029] The storage terminal provided by the present invention stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned power load prediction and intelligent suggestion method, and can achieve the technical effects of any of the above technical solutions, which will not be elaborated here.

[0030] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0032] Figure 1 One of the flowcharts of an embodiment of the power load prediction and intelligent suggestion method of the present invention is shown;

[0033] Figure 2 The diagram illustrates a step in determining a load prediction curve based on a time-series sequence of electricity load and a pre-set first prediction model in an embodiment of the present invention.

[0034] Figure 3 The diagram illustrates a step in determining a load forecast curve based on first load change forecast data and second load change forecast data in an embodiment of the electricity load forecasting and intelligent suggestion method of the present invention.

[0035] Figure 4 This diagram illustrates a process flow diagram of an embodiment of the electricity load forecasting and intelligent suggestion method of the present invention, prior to the step of determining the load forecast curve based on the electricity load time series and a pre-set first forecasting model.

[0036] Figure 5 This diagram illustrates a step in generating electricity consumption recommendations based on load forecasting curves, user behavior data, and electricity price information in an embodiment of the electricity load forecasting and intelligent recommendation method of the present invention.

[0037] Figure 6 The second schematic flowchart of an embodiment of the power load prediction and intelligent suggestion method of the present invention is shown.

[0038] Figure 7 A structural block diagram of a second power load prediction and intelligent suggestion system according to an embodiment of the present invention is shown;

[0039] Figure 8 A structural block diagram of an embodiment of the power load prediction and intelligent suggestion device of the present invention is shown;

[0040] Figure 9 A block diagram of an embodiment of the present invention, namely, a power load prediction and intelligent suggestion system, is shown. Detailed Implementation

[0041] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0043] Figure 1 One of the flowcharts of an embodiment of the electricity load forecasting and intelligent suggestion method of the present invention is shown, wherein the electricity load forecasting and intelligent suggestion method includes:

[0044] Step 102: Obtain historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information;

[0045] Step 104: Perform preprocessing operations on historical electricity consumption data, user behavior data, environmental data, and time data to obtain the electricity load time series sequence;

[0046] Step 106: Determine the load forecast curve based on the electricity load time series and the pre-set first forecast model, wherein the first forecast model is a hybrid model composed of variational mode decomposition model, long sequence time series forecast model and long short-term memory network model;

[0047] Step 108: Generate electricity consumption recommendations based on load forecast curves, user behavior data, and electricity price information.

[0048] The present invention provides a method for predicting and intelligently suggesting electricity loads. This method includes: first, acquiring historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information. Historical electricity consumption data can be obtained through smart meters; environmental data can include temperature or humidity; time data can include public holidays or weekends; user behavior data can include users' daily habits of starting and stopping electrical equipment and using electricity during fixed periods; and electricity price information refers to the electricity billing standards set by the power company in the user's region, such as peak-hour high-price electricity, off-peak low-price electricity, flat-rate base price, and tiered pricing rules. Subsequently, preprocessing operations are performed on historical electricity consumption data, user behavior data, environmental data, and time data to obtain the electricity load time series sequence. This sequence refers to a continuous sequence of electricity consumption values ​​arranged chronologically. Preprocessing involves integrating all scattered data at a unified time granularity, removing interfering temporary data, and standardizing data value ranges. This transforms the chaotic and irregular discrete electricity consumption data into structured time-series data that is continuously arranged in chronological order, formatted uniformly, and directly input into the model for computation. Following this, after obtaining the electricity load time series sequence, a load prediction curve is determined based on the sequence and a pre-set first prediction model. This first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model, and a long short-term memory network model. In other words, the first prediction model is an integrated prediction model that combines signal decomposition algorithms, long-sequence prediction models, and short-sequence prediction networks, operating synchronously in a dual-path manner. It can accurately predict both long-term electricity consumption trends and short-term fluctuations. After obtaining the first prediction model, the electricity load time series is input into it. Relying on the built-in dual-branch independent operation logic of the first prediction model, different variation patterns within the data are analyzed separately. The output results of the two operations are integrated to draw a continuous and complete long-term electricity load variation curve that closely reflects actual electricity consumption fluctuations. Finally, intuitive electricity fluctuation characteristics within the electricity load variation curve are extracted. Combined with user behavior data such as daily equipment start-up and shutdown, and fixed-time electricity consumption, and in accordance with electricity price information, targeted and actionable electricity consumption recommendations are formulated. This invention comprehensively covers electricity consumption influencing factors based on multi-dimensional basic data such as historical electricity consumption data, user behavior data, environmental data, and time data. Preprocessing operations eliminate computational errors caused by the aforementioned data deficiencies. Using the combined prediction model, i.e., the first prediction model, accurate predictions are achieved that take into account both long-term electricity consumption trends and short-term fluctuations. Finally, reasonable recommendations are matched based on actual usage scenarios. This achieves automated and accurate prediction of user electricity load, and at the same time, generates personalized electricity consumption guidance by combining users' actual electricity consumption habits with electricity price policies. This significantly reduces the workload of manual statistical analysis of electricity consumption data and improves the rationality of electricity planning.

[0049] Figure 2 The diagram illustrates a flowchart of a method for predicting and intelligently recommending electricity loads according to an embodiment of the present invention, specifically the step of determining a load prediction curve based on an electricity load time series and a pre-set first prediction model. The step of determining the load prediction curve based on the electricity load time series and the pre-set first prediction model includes:

[0050] Step 202: Determine the first mode component and the second mode component based on the power load time sequence and the variational mode decomposition model, wherein the frequency of the first mode component is greater than or equal to the preset frequency, and the frequency of the second mode component is less than the preset frequency;

[0051] Step 204: Determine the first load change prediction data based on the first modal component and the long short-term memory network model;

[0052] Step 206: Determine the second load change prediction data based on the second modal component and the long-sequence time-series prediction model;

[0053] Step 208: Determine the load forecast curve based on the first load change forecast data and the second load change forecast data.

[0054] In this embodiment, the step of determining the load prediction curve based on the electricity load time series and the first prediction model includes: firstly, inputting the electricity load time series into the variational mode decomposition model, whereby the variational mode decomposition model adaptively decomposes the original non-stationary, highly complex electricity load time series into a series of intrinsic mode components with different center frequencies and finite bandwidths. Subsequently, these intrinsic mode components are divided into two categories according to frequency: the first category is the first mode component, and the second category is the second mode component. The frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than the preset frequency. That is, the first mode component is a short-period, high-frequency, and highly volatile intrinsic mode component, while the second mode component is a long-period, low-frequency, and clearly trending intrinsic mode component. After obtaining the first and second modal components, the high-frequency first modal component is input separately into the Long Short-Term Memory (LSTM) network model. Relying on the gating structure within the LTM model, it focuses on learning short-term patterns such as sudden fluctuations in electricity consumption and temporary increases / decreases in electricity demand, outputting predictive data that accurately reflects short-term electricity consumption fluctuations—the first load change prediction data. Subsequently, the low-frequency second modal component is input into the long-sequence time-series prediction model. Relying on the sparse attention mechanism built into the long-sequence time-series prediction model, it mines stable electricity consumption cycle patterns and seasonal electricity consumption variation patterns over long periods, outputting predictive data that reflects the overall long-term electricity consumption trend—the second load change prediction data. Finally, the first and second load change prediction data are integrated to obtain the final load prediction curve. This invention utilizes a variational mode decomposition model to accurately decompose signal features, thereby separating different electricity consumption change patterns. Then, it matches the corresponding prediction model with the strongest adaptability to learn specific patterns, avoiding the shortcomings of a single model that cannot simultaneously take into account both long and short time series patterns. Finally, it integrates the two independent prediction results, taking into account both the global trend and local fluctuations, making the final prediction results more consistent with real electricity consumption scenarios.

[0055] Figure 3 This diagram illustrates a step in determining a load forecast curve based on first load change forecast data and second load change forecast data in an embodiment of the present invention's electricity load forecasting and intelligent suggestion method. The step of determining the load forecast curve based on the first load change forecast data and second load change forecast data includes:

[0056] Step 302: Perform weighted fusion and dimensional unification on the first load change prediction data and the second load change prediction data to obtain the third load change prediction data;

[0057] Step 304: Perform mapping calculations and numerical restoration on the third load change prediction data to obtain the load prediction curve.

[0058] In this embodiment, the step of determining the load forecast curve based on the first load change forecast data and the second load change forecast data includes: First, weighting and unifying the first load change forecast data and the second load change forecast data to obtain the third load change forecast data. That is, firstly, based on the proportion of the impact of long-term electricity consumption trends and short-term electricity fluctuations on the overall load, a differentiated fusion weight coefficient is set. Then, the first load change forecast data and the second load change forecast data are weighted and calculated according to the fusion weight coefficient. At the same time, the data dimension, data value format and time sampling interval of the first load change forecast data and the second load change forecast data are unified to eliminate the problems of inconsistent output data formats and misaligned time nodes of different models. After obtaining the third load change prediction data, the original electricity consumption data underwent normalization and data reduction in the previous preprocessing operations. Therefore, the fused third load change prediction data is not the actual value of the electricity load. At this point, a preset reverse mapping calculation formula is used to restore the standardized calculation data to the actual electricity consumption and power consumption values ​​that can be directly used in the real scenario. Then, based on the restored real continuous values, a visualized load prediction curve is generated according to the time sequence arrangement rules, and key peak and valley nodes of electricity consumption within the curve are marked. This invention distinguishes the degree of influence of different electricity consumption patterns through weighted fusion, ensuring that the integration results are clear in their primary and secondary aspects. It eliminates data interaction barriers by relying on dimensional unification, and then uses reverse numerical mapping rules to offset the numerical scaling effect caused by the previous data preprocessing operations, accurately restoring the actual electricity load values ​​and ensuring that the visualized curve data has practical application value.

[0059] Figure 4 The diagram illustrates a flowchart of an embodiment of the electricity load forecasting and intelligent suggestion method of the present invention, prior to the step of determining the load forecast curve based on the electricity load time series and a pre-set first prediction model. The method further includes the following steps before determining the load forecast curve based on the electricity load time series and the pre-set first prediction model:

[0060] Step 402: Determine the parameter combinations that need to be optimized among the variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model;

[0061] Step 404: Optimize the parameter combination according to the sparrow search algorithm to obtain the optimized parameter combination;

[0062] Step 406: Optimize the variational mode decomposition model, long sequence time series prediction model, and long short-term memory network model respectively based on the optimized parameter combination;

[0063] Step 408: Construct the first prediction model based on the optimized variational mode decomposition model, the optimized long sequence time prediction model, and the optimized long short-term memory network model.

[0064] In this embodiment, before determining the load forecast curve based on the electricity load time series and a pre-set first prediction model, the electricity load forecasting and intelligent suggestion method includes: First, pre-constructing the basic blank network architectures of the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model, respectively. Then, determining the parameter combinations that need optimization among the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model, wherein the parameter combinations include multiple parameters belonging to the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model. Subsequently, based on the parameter combinations, a sparrow search algorithm is used for continuous iterative updates, thereby globally selecting parameter combinations among the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model that are suitable for the current electricity consumption data characteristics. The sparrow search algorithm is an intelligent optimization algorithm that simulates the foraging and early warning escape behavior of sparrows in nature. After obtaining the optimized parameter combinations, the parameters from these combinations are respectively input into the basic blank network architectures of the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model to optimize these models. Finally, the optimized variational mode decomposition model, the optimized long-sequence time series prediction model, and the optimized long short-term memory network model are combined to obtain the first prediction model. This invention first establishes a standardized and general basic model architecture to ensure the overall framework's versatility. Then, relying on the powerful global search capability of the sparrow search algorithm, it quickly identifies a suitable configuration scheme within a massive range of parameter values. Finally, it combines the suitable configuration scheme with the basic architecture, allowing the final model to perfectly adapt to the characteristics of the electricity consumption time series data to be processed, thus ensuring model performance from both architectural and parameter perspectives.

[0065] In some embodiments, the parameter combination may optionally include: the number of modes of the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes of the long short-term memory network model, the number of network layers of the long short-term memory network model, the number of encoding layers of the long sequence time series prediction model, and the number of attention heads of the long sequence time series prediction model.

[0066] In this embodiment, the parameter combination may include: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time series prediction model, and the number of attention heads in the long sequence time series prediction model. By classifying the core parameters according to the three categories of variational mode decomposition model, long short-term memory network model, and long sequence time series prediction model, the key configuration items affecting the prediction accuracy and operation speed during the operation of the first prediction model are comprehensively covered. This allows the Sparrow Search algorithm to accurately focus on the core parameters to complete the optimization, ensuring that the final parameter combination can comprehensively optimize the overall performance of the model.

[0067] Figure 5 The diagram illustrates a flowchart of a method for generating electricity consumption recommendations based on load forecasting curves, user behavior data, and electricity price information, according to an embodiment of the present invention. The step of generating electricity consumption recommendations based on load forecasting curves, user behavior data, and electricity price information includes:

[0068] Step 502: Determine the peak and off-peak electricity consumption periods based on the load forecast curve;

[0069] Step 504: Determine users' electricity usage habits based on peak electricity consumption periods, off-peak electricity consumption periods, and user behavior data;

[0070] Step 506: Generate electricity usage suggestions based on users' electricity usage habits and electricity price information.

[0071] In this embodiment, the steps for generating electricity consumption recommendations based on load forecast curves, user behavior data, and electricity price information include: First, determining peak and off-peak electricity consumption periods based on the load forecast curve. That is, relying on the long-term load forecast curve, a unified load value threshold is set. Ranges with values ​​above the threshold are defined as peak consumption periods, and ranges with values ​​below the threshold are defined as off-peak periods. Simultaneously, the duration and peak load of peak and off-peak periods are statistically analyzed, and the distribution patterns of peak and off-peak electricity consumption periods on weekdays and weekends can be distinguished. Then, combining peak and off-peak periods with user behavior data, users' daily electricity consumption habits are determined. Subsequently, based on these habits, the defined peak and off-peak periods are matched, and users' regular electricity consumption habits, such as fixed high-power appliance usage periods, basic household electricity consumption periods, and idle / power-off periods, are summarized. Finally, referring to the electricity price information, i.e., the grid's peak-valley time-of-use pricing standard, and considering users' electricity consumption habits, targeted electricity consumption recommendations are given without significantly altering users' normal life and production electricity consumption order. This invention uses accurate load forecast curves as the basis for time patterns, real user electricity consumption data as the basis for usage habits, and official electricity pricing rules as the basis for economic regulation. The combination of these three data sources leads to an optimized electricity consumption plan that not only conforms to the actual electricity consumption situation but also aligns with electricity pricing policy guidelines, thus balancing practicality and economy.

[0072] In some embodiments, the preprocessing operations may optionally include one or a combination of missing value interpolation completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment and processing of multi-source data.

[0073] In this embodiment, preprocessing operations may include one or a combination of the following: missing value interpolation and completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment of multi-source data. Specifically, missing value interpolation and completion refers to filling in data gaps by extrapolating from adjacent normal data. Abnormal noise data removal refers to identifying and deleting erroneously collected data that deviates from the normal range of electricity consumption fluctuations. Multi-dimensional data normalization refers to converting data of different magnitudes, such as voltage, electricity consumption, temperature, and humidity, to the same numerical range. Dimensionality reduction of invalid and redundant features refers to deleting duplicate collected data that has no practical use in analyzing electricity consumption patterns, thus reducing the amount of data processing. Time-series alignment of multi-source data refers to unifying the collection time nodes of electricity consumption data, environmental data, and time data to achieve complete matching of the time axes of multiple data types.

[0074] In other words, when preprocessing historical electricity consumption data, user behavior data, environmental data, and time data, flexible combinations of processing methods can be selected based on the above data. For raw data with missing data, missing value interpolation is used to improve data continuity; for extreme erroneous data caused by collection failures, abnormal noise data is removed to purify the data; for multiple types of data with large differences in units and numerical ranges, multi-dimensional data normalization is used to unify data standards; when there is a large amount of repetitive and low-correlation invalid information within the data, invalid redundant feature dimensionality reduction is used to reduce the data volume; and for cases where the collection time of different types of data is not uniform, multi-source data time series alignment is used to unify the time benchmark. This invention comprehensively solves various common defects in raw collected electricity consumption data, such as data gaps, data errors, inconsistent standards, data redundancy, and time series misalignment, outputting high-quality, highly regularized time series data, reducing subsequent model prediction and calculation errors from the data source.

[0075] Figure 6 The second schematic flowchart illustrates an embodiment of the electricity load forecasting and intelligent suggestion method of the present invention, wherein the electricity load forecasting and intelligent suggestion method further includes:

[0076] Step 602: Obtain the user's historical electricity consumption data, user behavior data, environmental factors (such as temperature, humidity, etc.) and time factors (such as public holidays, weekends, etc.);

[0077] Step 604: Preprocess the acquired data by using linear interpolation to handle missing values, using quadrant plots to detect outliers and treating them as missing values, and finally using principal component analysis to reduce the dimensionality of the data.

[0078] Step 606: Input the preprocessed data into the variational mode decomposition model for data decomposition. Input the long-period, low-frequency, and trend-oriented components into the Informer model, while the short-period, high-frequency, and highly volatile components are used as the input of the long short-term memory network model. At the same time, use the sparrow search algorithm to search for the parameter combination of the VMD-Informer-LSTM parallel model. Finally, obtain the final prediction result through the fully connected layer and the output layer.

[0079] Step 608: Analyze the main driving causes of load peaks or changes based on load forecast results and user electricity consumption behavior;

[0080] Step 610: Based on the analysis results of load peak or change, provide smart electricity consumption suggestions in conjunction with time-of-use pricing.

[0081] In this embodiment, the electricity load forecasting and intelligent suggestion method mainly includes: First, acquiring users' historical electricity consumption data, user behavior data, environmental factors (such as temperature, humidity, etc.), and time factors (such as public holidays, public holidays, etc.) through data acquisition equipment; then, since the above data changes relatively smoothly and is continuous time series data, linear interpolation is used to perform linear calculations to fill in the missing values ​​between two adjacent valid values, so as to handle missing values; outlier detection is performed using the quadrant diagram method, and the detected outliers are treated as missing values, so that the subsequent prediction model focuses on learning the main load change patterns, thereby improving the stability and generalization ability of the prediction; after handling missing and outliers, principal component analysis (PCA) is used to reduce the data dimensionality, which improves the processing efficiency of subsequent prediction while retaining the highly correlated features in the original data. Subsequently, the preprocessed data is input into a Variational Mode Decomposition (VMD) model, adaptively decomposing the original non-stationary, highly complex load time series into a series of Intrinsic Mode Functions (IMFs) with different center frequencies and finite bandwidths. This method cleanly separates different components such as trends, daily cycles, and random noise, providing clearer and more targeted input features for subsequent prediction models. Each IMF component is thus more stable and regular than the original sequence, significantly reducing the learning difficulty for the subsequent Long Short-Term Memory (LSTM) network model.

[0082] Subsequently, the data after VMD decomposition was divided into two categories: long-period, low-frequency, and clearly trending components (such as weekly cycles, seasonal trends, and long-term trend terms) and short-period, high-frequency, and highly volatile components (such as daily cycles, random fluctuations, and high-frequency noise).

[0083] Long-term, low-frequency, and clearly trending components are input into the long-sequence time-series prediction (Informer) model. The Informer model utilizes its probabilistic sparse self-attention mechanism and attention distillation to efficiently uncover long-term global dependencies in the sequence.

[0084] The Informer model's probabilistic sparse self-attention mechanism selects the queries with the highest sparsity and only computes the attention between these selected queries and all keys. For unselected queries, their attention output is approximated by a mean vector. The following equation represents the output process of the probabilistic sparse self-attention mechanism:

[0085] ;

[0086] Where A represents the self-attention output feature matrix, and Q, K, and V are the query matrix, key matrix, and value matrix, respectively. Let d be the query matrix after probability sparse, d be the dimension of the input features, Softmax be the activation function, and T be the transpose of the matrix.

[0087] Attention distillation performs one-dimensional convolution and max pooling operations in the Informer model, halving the sequence length layer by layer to progressively extract trend features, strip away noise and short-term fluctuations, compress features into shorter sequences, and force the model to learn and retain the core long-term trends and cyclical patterns.

[0088] ;

[0089] in, Let be the input feature matrix of the (i+1)th multi-head sparse attention layer. Let be the output matrix of the i-th multi-head sparse attention layer, Convld be a one-dimensional convolution operation, ELU be the activation function, and MaxPool be the max pooling operation.

[0090] Short-cycle, high-frequency, and highly volatile components are input into LSTM. LSTM, through its unique gating mechanism (forget gate, input gate, and output gate) and cell state, captures these components with complex local temporal patterns, making it suitable for LSTM to capture short-term dependencies.

[0091] The current input of the LSTM and the hidden state of the previous time step are used as inputs to the forget gate, and the result is mapped to a value between 0 and 1 through the Sigmoid function. Values ​​close to 0 indicate that the corresponding cell state information will be forgotten, and values ​​close to 1 indicate that the information will be retained.

[0092] ;

[0093] Among them, W f and b f These are the weight matrix and bias vector of the forget gate, h, respectively. t-1 and x t Let σ be the hidden state from the previous time step and the input to the forget gate from the current time step. function, f t This is the output of the forget gate at the current moment. Therefore, the forget gate can determine the cell state C from the previous moment. t-1 How much important information will be retained?

[0094] The input gate determines how much of the current input information is updated in the cell state:

[0095] ;

[0096] ;

[0097] Among them, W i Let b be the weight matrix of the input gate. i W is the bias vector of the input gate. c Let b be the weight matrix of the candidate cell states. c h is the bias vector of the candidate cell state. t-1 and x t Let i be the hidden state from the previous time step and the input from the current time step, respectively. t The output of the input gate at the current moment. The candidate cell state represents the new information generated at the current moment, and tanh() represents the hyperbolic tangent activation function.

[0098] Based on the results of the forget gate and input gate, update the cell state:

[0099] ;

[0100] Among them, f t and i t The outputs of the forget gate and the input gate are respectively, C t-1 and C t These represent the cell states at the previous and current time points, respectively. This represents the current state of the candidate cells. This is the Hadamard product, which is a positional multiplication.

[0101] As can be seen from the above formula, the current cell state integrates new knowledge from both long-term and short-term memory.

[0102] The output gate determines which information from the cell state will be output as the hidden state at the current moment:

[0103] ;

[0104] ;

[0105] Among them, W o and b o These are the weight matrix and bias vector of the output gate, respectively, h t-1 Let x and y be the hidden state of the previous time step and the input of the current time step, respectively, and o be the hidden state of the previous time step and the input of the current time step. t C is the output of the output gate at the current moment. t h represents the current cell state. t This represents the current hidden state.

[0106] LSTM perfectly solves the three core contradictions in short-term prediction by combining gating mechanisms and cell state: long-term dependence and rapid response, regular patterns and random fluctuations, and individual differences and common characteristics.

[0107] To achieve optimal prediction, the Sparrow Search Algorithm (SSA) is introduced to find the parameter combination of the VMD-Informer-LSTM (Variational Mode Decomposition-Informer-Long Short-Term Memory) parallel model, thereby improving the accuracy and stability of short-term electricity load prediction.

[0108] The optimization algorithm draws inspiration from the foraging behavior and anti-predation mechanisms of sparrows. During foraging, this type of population is divided into three types: discoverers, joiners, and vigilants.

[0109] The discoverer is responsible for finding food and providing the population with foraging area information. After initializing the population and updating fitness values, the discoverer's location is updated.

[0110] ;

[0111] Where i is the number of the sparrow after sorting according to fitness in each iteration, that is, the i-th sparrow after sorting. Indicates the first The sparrow's position information in the j-th dimension during the t-th iteration. This represents the j-th dimension position information of the i-th sparrow in the (t+1)-th iteration, where α represents a random number in [0, 1], and iter max The maximum number of iterations is represented by Q, which is a random number following a normal distribution. L is a vector of all 1s. R2 represents the warning value, which ranges from [0, 1]. ST is the safety threshold, which ranges from [0.5, 1]. exp() represents the exponential function.

[0112] When the warning value is lower than the safety threshold, the search range is gradually narrowed and focused; otherwise, the strategy is randomly adjusted and the search moves to a safe area.

[0113] The participants use the discoverer to find food. The participants' positions are updated using different methods depending on their background.

[0114] ;

[0115] In the above formula, n is the sparrow population size, and Q is a random number that follows a normal distribution. This represents the worst position globally at the current moment. This represents the j-th dimension position information of the i-th sparrow in the t-th iteration, where i is the i-th sparrow after sorting. Let L be the position of the optimal discoverer at time t+1, and let L be an all-one vector; A + =A T (AA) T ) -1 For a matrix of shape 1×d, each element is randomly assigned a value. or d represents the number of parameters that the VMD-Informer-LSTM parallel model needs to find.

[0116] If a newcomer's fitness is in the top 50% of the population, a random position is found near the current optimal position, and the variance of each dimension from the optimal position will become smaller; otherwise, a new position is explored randomly. When the sparrow is a watcher, its position is updated according to the following formula.

[0117] ;

[0118] In the above formula, f i f is the fitness value of the current individual. g f is the current maximum fitness value. w This is the current minimum fitness value. and These represent the best and worst global positions at the current moment, respectively. Let represent the position information of the i-th sparrow in the j-th dimension during the t-th iteration, K be a random number in the range [-1, 1], β be a random number following a normal distribution with a mean of 0 and a variance of 1, and ε be the smallest constant chosen to avoid division by zero.

[0119] When the current sparrow is not the optimal individual (i.e., f) i >f g This indicates that the sparrow is at the edge of the population and needs to move to the globally optimal position. Move; and when the first sparrow is the optimal individual (i.e., f) i =f g If the search is in the vicinity of the current position, a fine search will be performed to avoid getting trapped in local optima.

[0120] In each iteration, the position is updated according to different formulas based on the different identities of each sparrow to find the optimal solution, i.e., the parameter combination.

[0121] The outputs of the Informer and LSTM branches are then fused through a fully connected layer and passed through an output layer to obtain the final load prediction result.

[0122] Subsequently, based on the predicted load curve, and combined with user electricity consumption behavior analysis, the peak values ​​and driving causes of load curve changes are identified, along with the specific electrical equipment causing these changes. Finally, based on the above load curve analysis results, and in conjunction with time-of-use pricing, intelligent electricity consumption suggestions are provided to achieve a better electricity consumption experience for users.

[0123] Figure 7 A structural block diagram of a second electricity load forecasting and intelligent suggestion system according to an embodiment of the present invention is shown. The present invention provides a second electricity load forecasting and intelligent suggestion system 70, including: a data acquisition module 702, a data preprocessing module 704, a load forecasting module 706, a load curve analysis module 708, and an intelligent electricity consumption suggestion module 710. The data acquisition module 702 acquires historical electricity consumption data, user behavior data, environmental factors (such as temperature and humidity), and time factors (such as public holidays and weekends). The data preprocessing module 704 preprocesses the data acquired by the data acquisition module 702, including: using linear interpolation to perform linear calculations to fill in missing values ​​between adjacent valid values; using quadrant plots to detect outliers and treating detected outliers as missing values; and using PCA to reduce data dimensionality after processing missing and outlier values. The load forecasting module 706 inputs the preprocessed data into VMD for signal decomposition, dividing the decomposed data into long-period, low-frequency, and clearly trending components and short-period, high-frequency, and highly volatile components, which serve as inputs to the Informer and LSTM models, respectively. Simultaneously, SSA is used to find the parameter combinations for the VMD-Informer-LSTM parallel model during the forecasting process. Finally, the outputs of the Informer and LSTM models are passed through a fully connected layer and an output layer to obtain the final load forecast result. The load curve analysis module 708 analyzes the peak value and driving factors of the predicted load curve based on user electricity consumption behavior. The smart electricity consumption suggestion module 710 makes intelligent decisions on electricity consumption planning based on the load curve analysis results and time-of-use pricing, and provides smart electricity consumption suggestions.

[0124] Figure 8 A structural block diagram of an electricity load forecasting and intelligent suggestion device according to an embodiment of the present invention is shown. The present invention provides an electricity load forecasting and intelligent suggestion device 80, including: a processor 802, a memory 804, a data acquisition device 806, and a communication unit 808. The data acquisition device 806 stores the acquired data and corresponding computer instructions for the aforementioned methods and systems in the memory 804 via the communication unit 808. The processor 802 executes these computer instructions to perform load forecasting and generate intelligent electricity consumption suggestions.

[0125] Figure 9 A structural block diagram of an embodiment of the present invention, namely an electricity load forecasting and intelligent suggestion system, is shown, wherein the electricity load forecasting and intelligent suggestion system 90 includes:

[0126] The first acquisition module 902 is used to acquire historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information;

[0127] The preprocessing module 904 is used to preprocess historical electricity consumption data, user behavior data, environmental data, and time data to obtain the electricity load time sequence.

[0128] The first determining module 906 is used to determine the load prediction curve based on the electricity load time series and the pre-set first prediction model, wherein the first prediction model is a hybrid model composed of variational mode decomposition model, long sequence time series prediction model and long short-term memory network model.

[0129] The generation module 908 is used to generate electricity consumption suggestions based on load forecast curves, user behavior data, and electricity price information.

[0130] The electricity load forecasting and intelligent suggestion system 90 provided by this invention mainly includes: a first acquisition module 902, a preprocessing module 904, a first determination module 906, and a generation module 908. The first acquisition module 902 first acquires historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information. Historical electricity consumption data can be obtained through smart meters; environmental data can be temperature or humidity, etc.; time data can be public holidays or weekends, etc.; user behavior data can include users' daily habits of starting and stopping electrical equipment and electricity consumption during fixed periods; and electricity price information refers to the electricity billing standards set by the power department in the user's area, such as peak-hour high-price electricity prices, off-peak low-price electricity prices, flat-rate base prices, and tiered pricing rules. Subsequently, the preprocessing module 904 performs preprocessing operations on historical electricity consumption data, user behavior data, environmental data, and time data to obtain an electricity load time series sequence. This electricity load time series sequence refers to a continuous sequence of electricity consumption values ​​arranged sequentially based on time. The preprocessing operation involves integrating all scattered data at a unified time granularity, eliminating temporary interference data, and standardizing the data value range. This transforms the chaotic and irregular discrete electricity consumption data into structured time-series data that is continuously arranged in chronological order, has a unified format, and can be directly input into the model for computation. Following this, after obtaining the electricity load time series sequence, the first determination module 906 determines the load prediction curve based on the electricity load time series sequence and a pre-set first prediction model. The first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model, and a long short-term memory network model. In other words, the first prediction model is an integrated prediction model that combines signal decomposition algorithms, long-sequence prediction models, and short-sequence prediction networks, operating synchronously in dual-path operation. It can accurately predict both long-term electricity consumption trends and short-term electricity fluctuations. After obtaining the first prediction model, the electricity load time series is input into it. Relying on the built-in dual-branch independent computation logic of the first prediction model, different variation patterns within the data are analyzed separately. The output results of the two computations are integrated to draw a continuous and complete long-term electricity load variation curve that closely reflects actual electricity consumption fluctuations. Finally, module 908 extracts intuitive electricity fluctuation characteristics from the electricity load variation curve. Combined with user behavior data such as daily start-up and shutdown of electrical equipment and electricity consumption during fixed periods, and in accordance with electricity price information, targeted and actionable electricity consumption recommendations are formulated. This invention comprehensively covers electricity consumption influencing factors based on multi-dimensional basic data such as historical electricity consumption data, user behavior data, environmental data, and time data. Preprocessing operations eliminate computational errors caused by the aforementioned data deficiencies. Using the combined prediction model, i.e., the first prediction model, accurate predictions that take into account both long-term electricity consumption trends and short-term electricity consumption fluctuations are achieved. Finally, reasonable recommendations are matched based on actual usage scenarios.It achieves automated and accurate prediction of user electricity load, and generates personalized electricity usage guidance by combining users' actual electricity usage habits with electricity pricing policies. This significantly reduces the workload of manual statistical analysis of electricity usage data and improves the rationality of electricity usage planning.

[0131] In some embodiments, optionally, the first determining module 906 is specifically configured to determine a first mode component and a second mode component based on the electricity load time series and the variational mode decomposition model, wherein the frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than the preset frequency; determine first load change prediction data based on the first mode component and the long short-term memory network model; determine second load change prediction data based on the second mode component and the long-sequence time series prediction model; and determine a load prediction curve based on the first load change prediction data and the second load change prediction data.

[0132] In this embodiment, the first determining module 906 is specifically used to first input the electricity load time series into the variational mode decomposition model. The variational mode decomposition model adaptively decomposes the original non-stationary, highly complex electricity load time series into a series of intrinsic mode components with different center frequencies and finite bandwidths. Then, these intrinsic mode components are divided into two categories according to frequency: the first category is the first mode component, and the second category is the second mode component. The frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than the preset frequency. That is, the first mode component is a short-period, high-frequency, and highly volatile intrinsic mode component, while the second mode component is a long-period, low-frequency, and clearly trend-oriented intrinsic mode component. After obtaining the first and second mode components, the high-frequency first mode component is separately input into the long short-term memory network model. Relying on the gating structure inside the long short-term memory network model, the model focuses on learning the short-term change patterns such as sudden fluctuations in electricity consumption and temporary increases and decreases in electricity consumption, and outputs prediction data that can accurately reflect short-term electricity consumption fluctuations, namely, the first load change prediction data. Subsequently, the second modal component of the low-frequency features is input into the long-sequence time-series prediction model. Relying on the sparse attention mechanism built into the long-sequence time-series prediction model, stable electricity consumption cycle patterns and seasonal electricity consumption variation patterns over long periods are mined, outputting prediction data that reflects the overall long-term electricity consumption trend—the second load change prediction data. Finally, the first and second load change prediction data are integrated to obtain the final load prediction curve. This invention utilizes a variational mode decomposition model to accurately decompose signal features, achieving separation of different electricity consumption variation patterns. Then, it matches the most suitable prediction model to learn specific patterns for each, avoiding the limitation of a single model that cannot simultaneously consider both long and short-sequence patterns. Finally, it integrates the two independent prediction results, taking into account both global trends and local fluctuations, making the final prediction result more closely reflect real-world electricity consumption scenarios.

[0133] In some embodiments, optionally, the first determining module 906 is specifically used to perform weighted fusion and dimensional unification on the first load change prediction data and the second load change prediction data to obtain the third load change prediction data; and to perform mapping calculation and numerical restoration on the third load change prediction data to obtain the load prediction curve.

[0134] In this embodiment, the first determining module 906 is specifically used to perform weighted fusion and dimension unification of the first load change prediction data and the second load change prediction data to obtain the third load change prediction data. That is, firstly, based on the proportion of the impact of long-term electricity consumption trends and short-term electricity fluctuations on the overall load, a differentiated fusion weight coefficient is set. Then, the first load change prediction data and the second load change prediction data are weighted and calculated according to the fusion weight coefficient. At the same time, the data dimension, data value format, and time sampling interval of the first load change prediction data and the second load change prediction data are unified to eliminate the problems of inconsistent output data formats and misaligned time nodes of different models. After obtaining the third load change prediction data, because the original electricity consumption data was normalized and reduced in the previous preprocessing operation, the third load change prediction data after fusion is not the actual value of the electricity load. At this time, through the preset reverse mapping calculation formula, the standardized calculation data is restored to the real electricity consumption and power consumption values ​​that can be directly used in the real scenario. Then, based on the restored real continuous values, a visual load prediction curve is generated according to the time sequence arrangement rules, and key electricity peak and valley nodes in the curve are marked. This invention distinguishes the degree of influence of different electricity consumption patterns through weighted fusion, ensuring that the integration results are clear in terms of primary and secondary factors. It eliminates data interaction barriers by relying on dimensional unification, and then uses reverse numerical mapping rules to offset the numerical scaling effect caused by the previous data preprocessing operations, accurately restoring the real electricity load values ​​and ensuring that the visualized curve data has practical application value.

[0135] In some embodiments, optionally, the electricity load forecasting and intelligent suggestion system 90 further includes: a second determining module, used to determine the parameter combination that needs to be optimized among the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model; a first optimizing module, used to optimize the parameter combination according to the sparrow search algorithm to obtain the optimized parameter combination; a second optimizing module, used to optimize the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model respectively according to the optimized parameter combination; and a construction module, used to construct a first prediction model according to the optimized variational mode decomposition model, the optimized long-sequence time-series prediction model, and the optimized long short-term memory network model.

[0136] In this embodiment, the electricity load forecasting and intelligent suggestion system 90 further includes a second determination module, a first optimization module, a second optimization module, and a construction module. First, the basic blank network architectures of the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model are pre-built. Then, the second determination module determines the parameter combinations that need optimization among the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model. These parameter combinations include multiple parameters belonging to each of the three models. Subsequently, the first optimization module iteratively updates the parameter combinations using a sparrow search algorithm, thereby globally selecting parameter combinations among the variational mode decomposition model, the long-sequence time-series prediction model, and the long short-term memory network model that are suitable for the current electricity load data characteristics. The sparrow search algorithm is an intelligent optimization algorithm that simulates the foraging and early warning escape behavior of sparrows in nature. After obtaining the optimized parameter combinations, the second optimization module inputs the parameters from the optimized parameter combinations into the basic blank network architectures of the variational mode decomposition model, the long-sequence time series prediction model, and the long short-term memory network model, respectively, to optimize these models. Finally, the construction module combines the optimized variational mode decomposition model, the optimized long-sequence time series prediction model, and the optimized long short-term memory network model to obtain the first prediction model. This invention first builds a standardized and general basic model architecture to ensure the general applicability of the overall model framework. Then, relying on the powerful global search capability of the sparrow search algorithm, it quickly identifies a suitable configuration scheme within a massive range of parameter values. Finally, it combines the suitable configuration scheme with the basic architecture, allowing the final model to perfectly adapt to the characteristics of the electricity consumption time series data to be processed, thus ensuring the model's performance from both architectural and parameter perspectives.

[0137] In some embodiments, the parameter combination may optionally include: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time series prediction model, and the number of attention heads in the long sequence time series prediction model.

[0138] In this embodiment, the parameter combination may include: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time series prediction model, and the number of attention heads in the long sequence time series prediction model, etc.

[0139] In some embodiments, optionally, the generation module 908 is specifically configured to determine peak and off-peak electricity consumption periods based on the load forecast curve; determine user electricity consumption habits based on peak and off-peak electricity consumption periods and user behavior data; and generate electricity consumption recommendations based on user electricity consumption habits and electricity price information.

[0140] In this embodiment, the generation module 908 is specifically used to determine peak and off-peak electricity consumption periods based on the load forecast curve. That is, relying on the long-term load forecast curve obtained from the forecast, a unified load value threshold is set. The intervals with values ​​higher than the threshold are defined as peak electricity consumption periods, and the intervals with values ​​lower than the threshold are defined as off-peak electricity consumption periods. Simultaneously, the duration and peak load of peak and off-peak periods are statistically analyzed, and the distribution patterns of peak and off-peak electricity consumption periods on weekdays and weekends can be distinguished. Subsequently, the user's daily electricity consumption habits are determined by combining peak and off-peak periods with user behavior data. Then, based on the user's daily electricity consumption habits, the defined peak and off-peak periods are matched, and the user's regular electricity consumption habits, such as fixed high-power appliance usage periods, basic household electricity consumption periods, and idle / power-off periods, are summarized. Finally, referring to the electricity price information, i.e., the grid's peak-valley time-of-use pricing standard, and combined with the user's electricity consumption habits, targeted electricity consumption suggestions are given without significantly altering the user's normal life and production electricity consumption order. This invention uses accurate load forecast curves as the basis for time patterns, real user electricity consumption data as the basis for usage habits, and official electricity pricing rules as the basis for economic regulation. The combination of these three data sources leads to an optimized electricity consumption plan that not only conforms to the actual electricity consumption situation but also aligns with electricity pricing policy guidelines, thus balancing practicality and economy.

[0141] In some embodiments, the preprocessing operations may optionally include one or a combination of missing value interpolation completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment and processing of multi-source data.

[0142] In this embodiment, the preprocessing operations may include one or a combination of the following: missing value interpolation completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment and processing of multi-source data.

[0143] One embodiment of the present invention provides an electricity meter, wherein the electricity meter includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the electricity load prediction and intelligent suggestion method as described above.

[0144] The electricity meter provided by the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described electricity load prediction and intelligent suggestion method, and can achieve the technical effects of any of the above embodiments, which will not be elaborated further.

[0145] One embodiment of the present invention provides a storage terminal on which a computer program is stored, which, when executed by a processor, implements the steps of the power load prediction and intelligent suggestion method as described above.

[0146] The storage terminal provided by the present invention stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-described method for predicting and intelligently suggesting electricity loads, and can achieve the technical effects of any of the above embodiments, which will not be elaborated further.

[0147] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance, unless otherwise expressly specified and limited. The terms "connection," "installation," and "fixing," etc., should be interpreted broadly. For example, "connection" can mean a fixed connection, a detachable connection, or an integral connection; it can mean a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0148] In the description of this specification, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0149] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting and intelligently recommending electricity load, characterized in that, include: Acquire historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information; The historical electricity consumption data, user behavior data, environmental data, and time data are preprocessed to obtain a time-series sequence of electricity load. The load prediction curve is determined based on the electricity load time series and a pre-set first prediction model, wherein the first prediction model is a hybrid model composed of variational mode decomposition model, long sequence time series prediction model and long short-term memory network model. Electricity consumption recommendations are generated based on the load forecast curve, the user behavior data, and the electricity price information. The step of determining the load forecast curve based on the electricity load time series and a pre-set first forecast model includes: The first mode component and the second mode component are determined based on the power load time sequence and the variational mode decomposition model, wherein the frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than the preset frequency; The first load change prediction data is determined based on the first modal component and the long short-term memory network model; The second load change prediction data is determined based on the second modal component and the long-sequence time-series prediction model; The load forecast curve is determined based on the first load change forecast data and the second load change forecast data; The step of determining the load forecast curve based on the first load change forecast data and the second load change forecast data includes: The first load change prediction data and the second load change prediction data are weighted, fused, and dimension-unified to obtain the third load change prediction data. The load forecast curve is obtained by mapping and numerical restoration of the third load change forecast data; Prior to the step of determining the load forecast curve based on the electricity load time series and the pre-set first forecast model, the electricity load forecasting and intelligent suggestion method further includes: Determine the parameter combinations that need to be optimized among the variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model; The parameter combination is optimized using the sparrow search algorithm to obtain the optimized parameter combination; The variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model are optimized according to the optimized parameter combination. The first prediction model is constructed based on the optimized variational mode decomposition model, the optimized long sequence time prediction model, and the optimized long short-term memory network model. The parameter combination includes: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time-series prediction model, and the number of attention heads in the long sequence time-series prediction model.

2. The method for predicting and intelligently recommending electricity load according to claim 1, characterized in that, Before the step of determining the parameter combination that needs to be optimized among the variational mode decomposition model, the long sequence time-series prediction model, and the long short-term memory network model, the electricity load forecasting and intelligent suggestion method further includes: The basic blank network architectures of the variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model are respectively constructed.

3. The method for predicting and intelligently recommending electricity load according to claim 1, characterized in that, The step of generating electricity consumption recommendations based on the load forecast curve, the user behavior data, and the electricity price information includes: The peak and off-peak electricity consumption periods are determined based on the load forecast curve. Determine user electricity usage habits based on the peak electricity consumption periods, the off-peak electricity consumption periods, and the user behavior data; The electricity usage suggestion is generated based on the user's electricity usage habits and the electricity price information.

4. The method for predicting and intelligently recommending electricity load according to any one of claims 1 to 3, characterized in that, The preprocessing operations include one or more of the following: missing value interpolation completion, abnormal noise data removal, multi-dimensional data normalization, dimensionality reduction of invalid and redundant features, and time-series alignment and processing of multi-source data.

5. A power load forecasting and intelligent suggestion system, characterized in that, include: The first acquisition module is used to acquire historical electricity consumption data, user behavior data, environmental data, time data, and electricity price information. The preprocessing module is used to perform preprocessing operations on the historical electricity consumption data, the user behavior data, the environmental data, and the time data to obtain the electricity load time series sequence. The first determining module is used to determine the load prediction curve based on the electricity load time series and a pre-set first prediction model, wherein the first prediction model is a hybrid model composed of a variational mode decomposition model, a long-sequence time series prediction model and a long short-term memory network model. The generation module is used to generate electricity consumption suggestions based on the load forecast curve, the user behavior data, and the electricity price information. The first determining module is specifically used to determine a first mode component and a second mode component based on the electricity load time series and the variational mode decomposition model, wherein the frequency of the first mode component is greater than or equal to a preset frequency, and the frequency of the second mode component is less than the preset frequency; determine first load change prediction data based on the first mode component and the long short-term memory network model; determine second load change prediction data based on the second mode component and the long-sequence time series prediction model; and determine the load prediction curve based on the first load change prediction data and the second load change prediction data. The first determining module is further specifically used to perform weighted fusion and dimension unification on the first load change prediction data and the second load change prediction data to obtain the third load change prediction data; The load forecast curve is obtained by mapping and numerical restoration of the third load change forecast data; The electricity load forecasting and intelligent suggestion system also includes: The second determining module is used to determine the parameter combination that needs to be optimized among the variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model. The first optimization module is used to optimize the parameter combination according to the sparrow search algorithm to obtain the optimized parameter combination. The second optimization module is used to optimize the variational mode decomposition model, the long sequence time prediction model, and the long short-term memory network model according to the optimized parameter combination. A construction module is used to construct the first prediction model based on the optimized variational mode decomposition model, the optimized long sequence time prediction model, and the optimized long short-term memory network model. The parameter combination includes: the number of modes in the variational mode decomposition model, the penalty coefficient of the variational mode decomposition model, the number of hidden layer nodes in the long short-term memory network model, the number of network layers in the long short-term memory network model, the number of encoding layers in the long sequence time-series prediction model, and the number of attention heads in the long sequence time-series prediction model.

6. An electricity meter, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power load forecasting and intelligent suggestion method as described in any one of claims 1 to 4.

7. A storage terminal storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the power load forecasting and intelligent suggestion method as described in any one of claims 1 to 4.

Images