Method for controlling a light storage and charging energy unit, program product, medium and device

Abstract

The application provides a kind of light storage energy charging unit control method, program product, medium and equipment, it is related to power supply field.Therein above-mentioned control method includes: the electric energy historical data of light storage energy charging unit is collected, and the electric energy influence factor information corresponding to electric energy historical data is obtained;Electric energy historical data and influence factor information are processed into the multidimensional data set of uniform data format;From multidimensional data set, the core feature set is extracted;Light storage energy charging unit is clustered using core feature set, and the unit grouping similar to electric energy application behavior is obtained;Light storage energy charging unit is controlled using the AI control model corresponding to unit grouping.The scheme of the application can comprehensively consider various factors to control the electric energy use condition of light storage energy charging unit to improve the adaptability and effectiveness of overall control.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power supply, in particular to a control method, program product, medium and equipment of a light storage and charging energy unit. BACKGROUND

[0002] The light storage and charging energy unit is a comprehensive energy solution integrating photovoltaic power generation, energy storage system and charging facilities. For example, household and small commercial power users can be independent light storage and charging energy units after installing photovoltaic and energy storage devices. The light storage and charging energy unit has the characteristics of large number, large scale and great difference.

[0003] In the face of the above challenges, a more effective control method of the light storage and charging energy unit is needed to regulate power supply and power consumption, meet the power demand of users and reduce the impact on the power grid.

[0004] Compared with the control of large-scale power grid, the power use characteristics of the light storage and charging energy unit have the characteristics of great difference in user power use behavior, many external factor disturbances, complex time sequence rules and insufficient prediction reliability, which makes the existing system-level power grid operation control method not directly applicable. Therefore, the existing control method of the light storage and charging energy unit cannot adapt to the characteristics of users when using a general model for prediction and control, resulting in a high failure rate of control. The light storage and charging energy unit has limited data and poor training convergence, and the model accuracy is limited, which leads to the lack of an accurate and efficient method for the light storage and charging energy unit in the prior art. SUMMARY

[0005] An object of the present application is to provide a control method of a light storage and charging energy unit based on an AI model, which at least solves any of the above technical problems.

[0006] A further object of the present application is to solve the problem of great difference in user power use behavior.

[0007] A further object of the present application is to enhance the foresight and accuracy of control.

[0008] In particular, the present application provides a control method of a light storage and charging energy unit based on an AI model. The method comprises:

[0009] Collecting power historical data of the light storage and charging energy unit and obtaining power influence factor information corresponding to the power historical data;

[0010] processing the electric energy historical data and the influencing factor information into a multi-dimensional data set with uniform data format;

[0011] extracting a core feature set from the multi-dimensional data set;

[0012] clustering the light storage and charging energy units using the core feature set to obtain unit groups with similar electric energy application behaviors;

[0013] controlling the light storage and charging energy units using an AI control model corresponding to the unit groups.

[0014] Optionally, the step of controlling the light storage and charging energy units using the AI control model corresponding to the unit groups comprises:

[0015] obtaining a time series prediction network model and an AI control model pre-constructed for each unit group;

[0016] predicting real-time data of the light storage and charging energy units using the time series prediction network model to obtain electric energy prediction data;

[0017] generating a control instruction from the AI control model according to the real-time data and the electric energy prediction data, and executing the control instruction by the light storage and charging energy units.

[0018] Optionally, the step of predicting the real-time data of the light storage and charging energy units using the time series prediction network model comprises:

[0019] collecting real-time data of the light storage and charging energy units, the real-time data including electric energy real-time data and real-time electric energy influencing factor information;

[0020] determining a unit group to which the light storage and charging energy unit belongs;

[0021] predicting the electric energy real-time data and the real-time electric energy influencing factor information using a time series prediction network model of the unit group to which the light storage and charging energy unit belongs to obtain electric energy prediction data.

[0022] Optionally, the step of collecting electric energy historical data of the light storage and charging energy unit comprises: collecting original electric energy data from a smart meter terminal and a power supply and consumption terminal of the light storage and charging energy unit, and performing numerical processing on the original electric energy data according to a preset sampling period to obtain the electric energy historical data;

[0023] The step of obtaining the electric energy influencing factor information comprises: obtaining meteorological data from an environmental data management platform, and determining date marker data corresponding to the meteorological data according to a calendar database, and performing normalization processing on the meteorological data and the date marker data to obtain the electric energy influencing factor information.

[0024] Optionally, the step of processing the electric energy historical data and the electric energy influencing factor information into a multi-dimensional data set with uniform data format comprises:

[0025] The power historical data and the power influencing factor information are time-aligned to ensure that they correspond in the time dimension;

[0026] The power historical data and the power influencing factor information after the time alignment are converted according to a preset data format;

[0027] The power historical data and the power influencing factor information after the data format conversion are arranged into a multidimensional data set.

[0028] Optionally, the step of extracting the core feature set from the multidimensional data set comprises:

[0029] The quantile autocovariance of the power historical data in the multidimensional data set is calculated to generate a plurality of behavior features reflecting the fluctuation law of the power historical data;

[0030] The power influencing factor information in the multidimensional data set is feature-encoded to generate external influence features;

[0031] The behavior features and the external influence features are combined to form multidimensional initial features;

[0032] The multidimensional initial features are dimension-reduced to obtain the core feature set.

[0033] Optionally, the step of dimension-reducing the multidimensional initial features comprises:

[0034] The covariance matrix of the multidimensional initial features is calculated;

[0035] The eigenvalues of the covariance matrix and the corresponding orthonormal unit eigenvectors are solved;

[0036] The dimensions of the multidimensional initial features are sorted according to the eigenvalues;

[0037] The features corresponding to the eigenvalues with high ranks are extracted to form the dimension-reduced multidimensional initial features.

[0038] Optionally, the step of extracting the features corresponding to the eigenvalues with high ranks comprises:

[0039] The cumulative variance contribution rate is calculated in sequence according to the order of the eigenvalues;

[0040] The dimension of the extracted features is determined according to the size of the cumulative variance contribution rate.

[0041] Optionally, the step of clustering the optical storage and charging energy units using the core feature set comprises:

[0042] The similarity between the core feature sets corresponding to the optical storage and charging energy units is calculated;

[0043] merge the light storage and charging energy units into a plurality of initial unit groups according to the similarity using a hierarchical clustering method;

[0044] Optimize the fuzzy mean clustering optimized by the genetic algorithm for the plurality of initial unit groups to obtain unit groups with similar electric energy application behaviors.

[0045] According to another aspect of the present application, a computer program product is also provided, which comprises a computer program that, when executed by a processor, implements the steps of any of the above AI model-based control methods for light storage and charging energy units.

[0046] According to still another aspect of the present application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of any of the above AI model-based control methods for light storage and charging energy units.

[0047] According to still another aspect of the present application, a computer device is also provided, which comprises a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of any of the above AI model-based control methods for light storage and charging energy units.

[0048] The AI model-based control method for light storage and charging energy units of the present application processes electric energy historical data and electric energy influencing factor information into a multi-dimensional data set and extracts a core feature set, then clusters, and constructs a prediction model according to the class. Unit groups with similar electric energy application behaviors are controlled using corresponding AI control models. The AI control model aggregates electric energy data of a large number of users in the group, which can avoid the problem of poor model convergence caused by insufficient electric energy data of a single user. In addition, the electric energy application behaviors in the same group are similar, which can avoid the problem of not meeting the characteristics of users caused by large differences in electric energy use behaviors of users. Therefore, the electric energy use of the light storage and charging energy unit can be controlled by comprehensively considering various factors, which provides a scientific basis for user energy saving guidance and improves the accuracy and effectiveness of electric energy management. This scheme accurately captures the differences in electric energy application behaviors between units, makes the control more suitable for the characteristics of each group, and improves the adaptability and effectiveness of overall control.

[0049] Further, the scheme of the present application collects raw electric energy data from a specific terminal and processes them according to a preset period, and obtains electric energy influencing factor information from a specific platform and normalizes the information, so that the obtained data is more standardized and usable, which is beneficial to subsequent accurate analysis and modeling and improves the data processing quality.

[0050] Further, the scheme of the present application performs covariance matrix-based dimension reduction operation on multi-dimensional initial features, reasonably reduces feature dimension from a mathematical principle, reduces calculation complexity while retaining main information, and improves subsequent processing efficiency.

[0051] Further, the scheme of the present application trains and optimizes using a random gradient descent algorithm, utilizes the processing advantage of LSTM on time series data, optimizes training to improve the performance of the prediction model, and thus improves prediction accuracy.

[0052] The above and other objects, advantages and features of the present application will become more apparent from the following detailed description of some embodiments thereof, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0053] Some embodiments of the present application will be described in detail with reference to the drawings, wherein the same or like reference numerals that denote the same or similar components or parts are used throughout the various drawings. It should be understood by those skilled in the art that the drawings are not necessarily to scale. In the drawings:

[0054] Figure 1 is a schematic diagram of a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0055] Figure 2 is a schematic diagram of processing a multi-dimensional data set of uniform data format in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0056] Figure 3 is a schematic diagram of processing a multi-dimensional data set of uniform data format in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0057] Figure 4 is a schematic diagram of performing dimension reduction on multi-dimensional initial features in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0058] Figure 5 is a schematic diagram of using a core feature set to cluster an optical storage and charging energy unit in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0059] Figure 6 is a schematic diagram of constructing a time series prediction network model and applying it in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application;

[0060] Figure 7is a schematic diagram of a computer program product according to an embodiment of the present application;

[0061] Figure 8 is a schematic diagram of a computer-readable storage medium according to an embodiment of the present application;

[0062] Figure 9 is a schematic block diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0063] Those skilled in the art will understand that the embodiments described below are merely a part of the embodiments of the present application, rather than all the embodiments of the present application, and are intended to explain the technical principles of the present application, rather than limit the protection scope of the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort shall fall within the protection scope of the present application.

[0064] The embodiment of the present application provides a control method of a light-storage-charging energy unit based on an AI model, which is used for analyzing an electric energy application state of the light-storage-charging energy unit. The light-storage-charging energy unit is an independent user with independent electric energy metering requirements, and is generally a low-voltage power distribution network user. The electric energy data generally includes electric energy data (such as electric energy consumption data, load power data), photovoltaic power generation data (such as power generation power data, power generation data) and energy storage charging and discharging data (such as charging and discharging power, charging and discharging state, charging and discharging period, etc.).

[0065] Figure 1 is a schematic diagram of a control method of a light-storage-charging energy unit based on an AI model according to an embodiment of the present application, which can include:

[0066] In step S101, electric energy historical data of the light-storage-charging energy unit is collected, and electric energy influence factor information corresponding to the electric energy historical data is obtained.

[0067] The above-mentioned step of collecting the electric energy historical data of the light-storage-charging energy unit includes: collecting original electric energy data from a smart meter terminal of the light-storage-charging energy unit and a power supply and use equipment terminal (such as an electric load equipment, a photovoltaic control equipment, an energy storage control equipment, etc.), performing numerical processing on the original electric energy data according to a preset sampling period, and obtaining the electric energy historical data. The electric energy historical data can be collected by communicating with the smart meter, the power supply and use terminal (for example, a load equipment with electric energy metering or electric signal measurement function, a control equipment of a light-storage-charging equipment). The electric energy historical data can include: user number, collection time, electric energy consumption, power generation, power, on-off state data of electric equipment, photovoltaic operation data, energy storage charging and discharging state data, current value, voltage value, harmonic content rate, etc.

[0068] Alternatively, the electricity consumption history data can also be obtained from a data server of a power company, a load device, or a light storage device.

[0069] The step of obtaining electricity consumption influencing factor information can include: obtaining meteorological data from an environmental data management platform, determining date marker data corresponding to the meteorological data according to a calendar database, and performing normalization processing on the meteorological data and the date marker data to obtain the electricity consumption influencing factor information. The electricity consumption influencing factor information can be obtained by communicating with devices such as an environmental data management platform (e.g., a meteorological platform, etc.), a calendar database, etc. The electricity consumption influencing factor information can include meteorological data (daily maximum temperature, minimum temperature, average temperature, humidity, precipitation, weather type (sunny / cloudy / rain / snow, etc.)), date data (month, day of the week, whether it is a holiday, etc.). These electricity consumption influencing factor information will have an impact on the electricity consumption of the light storage and charging energy unit. In addition to meteorological and date marker data, the electricity consumption influencing factor information can also consider factors such as commercial activities around the community.

[0070] Step S102, processing the electricity consumption history data and the electricity consumption influencing factor information into a multi-dimensional data set with unified data format.

[0071] Step S103, extracting a core feature set from the multi-dimensional data set.

[0072] Step S104, using the core feature set to cluster the light storage and charging energy units to obtain unit groups with similar electricity consumption behaviors.

[0073] Step S105, using an AI control model corresponding to the unit groups to control the light storage and charging energy units. One optional control method is: obtaining a time series prediction network model and an AI control model pre-constructed for each unit group; using the time series prediction network model to predict real-time data of the light storage and charging energy units to obtain electricity prediction data; using the AI control model to generate control instructions according to the real-time data and the electricity prediction data, and executing the control instructions by the light storage and charging energy units.

[0074] The control method of the AI model-based optical storage and charging energy unit of the above embodiment processes the electric energy historical data and the electric energy influencing factor information into a multi-dimensional data set and extracts a core feature set, and then clusters and constructs a prediction model according to the class. The units with similar electric energy application behaviors are grouped to construct a prediction model and an AI control model, respectively. The electric energy data of a large number of users in the group can avoid the problem of poor model convergence caused by insufficient electric energy data of a single user. In addition, the electric energy application behaviors in the same group are similar, which can avoid the problem of not meeting the user characteristics caused by large differences in user electric energy use behaviors. Therefore, the electric energy use of the optical storage and charging energy unit can be classified and predicted by comprehensively considering various factors, which can provide a scientific basis for electric energy use regulation, user energy saving guidance, and the like, and improve the accuracy and effectiveness of electric energy management.

[0075] Figure 2 FIG. 1 is a schematic diagram of processing a multi-dimensional data set with a unified data format according to an AI model-based optical storage and charging energy unit control method according to an embodiment of the present application. The step of processing the electric energy historical data and the electric energy influencing factor information into a multi-dimensional data set with a unified data format can include:

[0076] In step S201, the electric energy historical data and the electric energy influencing factor information are time-aligned to ensure that they correspond in the time dimension.

[0077] The time alignment of the electric energy historical data and the electric energy influencing factor information can be performed according to the collection period for missing value processing and outlier processing. The missing value processing can be performed by using an average value smoothing method to complete the missing value, for example, by taking the average value of a plurality of sampling points before and after the missing value as the completion value of the missing value.

[0078] Outliers can be identified by numerical anomalies, such as a sudden increase in load value to a preset multiple, or a negative humidity value, a value significantly exceeding the limit, etc. Outliers can be directly discarded and completed in a similar manner to missing values.

[0079] Continuous power values and the like can be normalized to the [0, 1] interval using min-max normalization to avoid magnitude differences that interfere with model training, i.e., the maximum power value is 1, and the minimum value is 0. Then, the data between the minimum value and the maximum value is converted to the [0, 1] interval to eliminate the influence of the absolute size of the value.

[0080] In step S202, the electric energy historical data and the electric energy influencing factor information processed by the time alignment are converted according to a predetermined data format. The data format can be defined according to empirical values to select a format that meets the accuracy requirements and eliminates the influence of the dimension.

[0081] Step S203, the electric energy historical data and the electric energy influencing factor information after data format conversion are arranged into a multidimensional data set.

[0082] The above process makes the obtained data more standardized and available, facilitates subsequent accurate analysis and modeling, and improves data processing quality.

[0083] Figure 3 is a schematic diagram of processing a multidimensional data set of uniform data format in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application. The step of extracting a core feature set from the multidimensional data set can include:

[0084] Step S301, calculate the quantile autocovariance (QC) of the electric energy historical data in the multidimensional data set, thereby generating a plurality of behavior characteristics reflecting the fluctuation law of the electric energy historical data. The calculation process can include: calculating the optimal value of a plurality of key quantiles; calculating a plurality of quantiles using a pinball loss function, and obtaining behavior characteristics through quantile autocovariance, which can reflect the fluctuation law of the electric energy historical data. Given the time series of the electric energy historical data as {X1, …, Xn}, the QC calculation expression is as follows:

[0085] , in the formula is the autocovariance of the time series under the quantile τ∈(0, 1) and , and j is the time delay.

[0086] is the optimal conditional value q τ corresponding to the characteristic function.

[0087] Step S302, feature encoding is performed on the electric energy influencing factor information in the multidimensional data set to generate external influence characteristics;

[0088] Step S303, the behavior characteristics and the external influence characteristics are combined to form multidimensional initial characteristics; the multidimensional initial characteristics include electric energy historical data characteristics and external influence characteristics.

[0089] Step S304, dimension reduction is performed on the multidimensional initial characteristics to obtain a core feature set.

[0090] The above process generates behavior characteristics by calculating quantile autocovariance, encodes, combines and reduces the electric energy influencing factor information, extracts a more representative core feature set, reduces data redundancy, highlights key features, and improves the accuracy of clustering and prediction.

[0091] The dimension reduction can use a PCA (Principal Components Analysis) dimension reduction algorithm. Figure 4 is a schematic diagram of dimension reduction on multi-dimensional initial features in a control method of an AI model-based optical storage and charging energy unit according to an embodiment of the present application. The step of dimension reduction on multi-dimensional initial features can include:

[0092] In step S401, a covariance matrix of the multi-dimensional initial features is calculated.

[0093] In step S402, eigenvalues of the covariance matrix and corresponding orthonormal unit eigenvectors are solved.

[0094] In step S403, dimensions of the multi-dimensional initial features are sorted according to the eigenvalues. The sorting manner is to arrange in descending order.

[0095] In step S406, features corresponding to the top-ranked eigenvalues are extracted to form the multi-dimensional initial features after dimension reduction.

[0096] The step of extracting features corresponding to the top-ranked eigenvalues can further include:

[0097] In step S404, cumulative variance contribution rates are calculated in sequence according to the order of the eigenvalues. That is, the cumulative calculation of the variance contribution rate is performed one by one according to the sorting.

[0098] In step S405, the dimension of the extracted features is determined according to the size of the cumulative variance contribution rate. That is, the dimension whose cumulative variance contribution rate is greater than or equal to a set threshold (for example, 85%, 90%) is retained, so that neither the individual differences of user behavior nor the influence of external factors on the load is ignored. For example, if the cumulative variance contribution rate of the first r influencing factors is greater than or equal to the set threshold, then r is taken as the dimension of the extracted features, that is, r-dimensional features are retained.

[0099] The step of calculating the quantile autocovariance of the power history data to generate the behavior features can effectively capture the fluctuation rules of the power history data, providing key information for analyzing the household power application behavior. The feature encoding of the power influencing factor information generates external influence features, which are combined with the behavior features to form multi-dimensional initial features, comprehensively considering the internal power application behavior and external influencing factors, making the data features more comprehensive. The dimension reduction on the multi-dimensional initial features obtains the core feature set, removes redundant information, highlights key features, and improves the subsequent analysis efficiency and model accuracy.

[0100] The multi-dimensional initial features are sorted according to eigenvalues and the features corresponding to the front eigenvalues are extracted for dimension reduction, so that the feature dimension is reasonably reduced, the calculation complexity is reduced on the premise of retaining main information, subsequent processing is more efficient, and the generalization ability of the model is improved.

[0101] The cumulative variance contribution rate is calculated according to the eigenvalue sequence, and the dimension of the extracted features is determined according to the size, thereby providing a scientific quantitative basis for the dimension reduction operation, avoiding the blindness of feature extraction, and optimizing the feature set while retaining data information to further improve the accuracy and effectiveness of data analysis.

[0102] Figure 5 is a schematic diagram of using a core feature set to cluster the light storage and charging energy unit according to an embodiment of the application. The step of using the core feature set to cluster the light storage and charging energy unit can include:

[0103] Step S501, the similarity between the core feature sets corresponding to the light storage and charging energy unit is calculated. The similarity can be measured by using the Euclidean distance.

[0104] Step S502, the light storage and charging energy unit is merged into multiple initial unit groups according to the similarity using a hierarchical clustering method. The hierarchical clustering method first regards each user as a separate cluster, and then gradually merges the nearest clusters. In this process, the "silhouette coefficient" is used to judge the clustering effect, and the cluster number when the silhouette coefficient is maximum is taken as the initial unit group.

[0105] The silhouette coefficient (Silhouette Coefficient, abbreviated as SC) is a way to evaluate the clustering result, which reflects the clustering effect in terms of intra-class aggregation and inter-class separation. The SC calculation expression can be:

[0106] SC(i)=[u(i)-v(i)] / max(u(i),v(i)), where SC(i) takes a value range of [-1, 1], SC(i)=1 indicates that the intra-class aggregation and inter-class separation are both optimal. u(i) is the intra-class aggregation, which represents the average distance between unit i and all units in the class. v(i) is the inter-class separation, which represents the average distance between unit i and all units in the nearest adjacent class. The average value of the silhouette coefficients of all units is the silhouette coefficient of the clustering result.

[0107] Step S503, the fuzzy mean clustering optimized by the genetic algorithm is used to optimize the plurality of initial unit groups, and unit groups with similar electric energy application behaviors are obtained. The genetic algorithm optimization can combine a genetic algorithm (GA) and a fuzzy C mean clustering (FCM). The optimization process is as follows: first, initialize the population, and randomly generate a set of candidate clustering centers as the initial population; then, evaluate the fitness, perform clustering on each candidate clustering center using the FCM algorithm, and calculate the fitness (such as the segmentation accuracy or the silhouette coefficient); then, perform the selection operation, select the optimal candidate clustering center according to the fitness, and retain high-quality solutions; then, perform the crossover and mutation, perform the crossover and mutation operation on the selected candidate clustering center, generate a new candidate solution, and thus expand the search range; finally, perform the iterative optimization, repeat the evaluation, selection, crossover and mutation steps until the termination condition (such as the maximum number of iterations or the convergence of the fitness) is met.

[0108] The above process calculates the similarity between the core feature sets of the light storage and charging energy units, and uses a hierarchical clustering method to obtain a plurality of initial unit groups based on the similarity, to preliminarily gather users with similar electric energy application behaviors together. Then, the fuzzy mean clustering optimized by the genetic algorithm is used to optimize the initial groups, which can more accurately mine the similarity between users and obtain more accurate unit groups with similar electric energy application behaviors, to provide a basis for targeted electric energy use prediction and personalized services.

[0109] Figure 6 FIG. 1 is a schematic diagram of constructing a time series prediction network model and applying the same in an AI model-based control method of a light storage and charging energy unit according to an embodiment of the present application. The step of constructing a time series prediction network model for each unit group can include:

[0110] Step S601, LSTM time series model training is performed using the core feature set of each unit group;

[0111] Step S602, the LSTM time series model is optimized and trained using a stochastic gradient descent algorithm, to obtain a time series prediction network model.

[0112] After the step of constructing a time series prediction network model for each unit group, the method of the present embodiment further includes:

[0113] Step S603, electric energy real-time data of the light storage and charging energy unit are collected, and real-time electric energy influencing factor information is obtained;

[0114] Step S604, the unit group to which the light storage and charging energy unit belongs is determined;

[0115] Step S605, the time series prediction network model of the unit group to which the light storage and charging energy unit belongs is used to predict the electric energy real-time data and the real-time electric energy influencing factor information, to obtain electric energy prediction data.

[0116] At step S606, the AI control model generates control instructions based on real-time data and electricity prediction data, and the light storage and charging energy unit executes the control instructions.

[0117] The above model construction process trains an LSTM time series model for each unit group using a core feature set, fully leveraging the processing advantages of LSTM for time series data, and effectively learning the regularity of electricity use in the time dimension of the light storage and charging energy unit. Using a stochastic gradient descent algorithm to optimize and train the LSTM time series model can speed up model convergence, improve model performance and prediction accuracy, and thus provide more reliable electricity prediction data for the light storage and charging energy unit in the unit group.

[0118] The AI control model calculates a set of optimal control parameters based on real-time data and electricity prediction data according to the control characteristics of the corresponding unit group (such as economic control preference, system stability and safety preference, grid dependence preference, etc.), and formulates a control instruction sequence. The control instruction sequence is issued to the local controllers (such as photovoltaic inverters, energy storage converter PCS, battery management system BMS, charging pile controller, and electrical equipment controller) of each device through Internet of Things (IoT) communication protocols (such as MQTT, Modbus, CAN bus, etc.). After receiving the instructions, each local controller will parse and execute the corresponding operations, such as adjusting power and switching switch states. The AI control model can continuously monitor the effects after the instructions are executed, and use these real-time data as the basis for subsequent prediction and control, forming a control loop, so that the AI model can continuously learn and optimize its decision-making strategy to adapt to changing environments and device states.

[0119] The above model uses the process to collect real-time electricity data and real-time electricity influencing factor information of the light storage and charging energy unit, determines the user's group, and uses the time series prediction network model of the corresponding group for prediction, which can timely and accurately obtain the future electricity use of the light storage and charging energy unit, provide real-time data support for real-time power dispatching and user real-time energy saving suggestions, and enhance the practicality and timeliness of electricity data analysis.

[0120] The embodiment also provides a computer program product 810, a computer readable storage medium 820, and a computer device 830. Figure 7 is a schematic diagram of the computer program product 810 according to an embodiment of the present application, Figure 8 is a schematic diagram of the computer readable storage medium 820 according to an embodiment of the present application, Figure 9 is a schematic block diagram of the computer device 830 according to an embodiment of the present application.

[0121] The computer program product 810 includes a computer program 811, which when executed by the processor 831 implements the steps of any of the above-described AI model-based control methods of the optical storage and charging energy unit. The computer-readable storage medium 820 has the above-described computer program 811 stored thereon, which when executed by the processor 831 implements the steps of any of the above-described AI model-based control methods of the optical storage and charging energy unit. The computer device 830 can include a memory 832, a processor 831, and a computer program 811 stored on the memory 832 and running on the processor 831.

[0122] The computer program 811 for performing the operations of the present application can be in an assembly language, an Instruction Set Architecture (ISA) language, machine language, a machine dependent instruction, a microcode, a firmware instruction, a state setting data, a configuration data of an integrated circuit, or source code or object code written in any combination of one or more programming languages and process programming languages.

[0123] The computer program 811 can be executed in whole or in part on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or connected to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, to perform aspects of the present application, electronic circuits, including for example programmable logic circuitry, Field-Programmable Gate Arrays (FPGA) or Programmable Logic Arrays (PLA), can be personalized by utilizing state information of the computer readable program instructions to perform the computer readable program instructions to personalize the electronic circuits.

[0124] For the description of the present embodiment, the computer program product 810 is a relevant product containing the computer program 811.

[0125] For the purposes of this description, a computer readable storage medium 820 is a tangible device that can retain and store computer- programmable instructions 811. The computer readable storage medium 820 can be any available medium or device that can be accessed by a computer. By way of example, and not limitation, such computer readable storage medium 820 can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium / mode of mass storage deployment or technology. Computer readable storage medium 820 can include a computer readable storage medium 820 having one or more computer readable storage media 820. The computer readable storage medium 820 can also be computer readable communication medium 830.

[0126] To thus far, those skilled in the art will appreciate that although exemplary embodiments of the application have been shown and described herein, various modifications and changes can be made without departing from the spirit and scope of the application. Accordingly, all such modifications and changes are intended to be included within the scope of the application.

Claims

1. A control method of an AI model-based optical storage and charging energy unit, characterized by The method comprises the following steps: Collecting electric energy historical data of the light storage and charging energy unit and obtaining electric energy influencing factor information corresponding to the electric energy historical data; Processing the electric energy historical data and the influencing factor information into a multi-dimensional data set with unified data format; Extracting a core feature set from the multi-dimensional data set; Clustering the light storage and charging energy unit using the core feature set to obtain a unit grouping with similar electric energy application behaviors; Controlling the light storage and charging energy unit using an AI control model corresponding to the unit grouping, wherein the step of extracting the core feature set from the multi-dimensional data set comprises: Calculating the quantile autocovariance of the electric energy historical data in the multi-dimensional data set to generate a plurality of behavior features reflecting the fluctuation law of the electric energy historical data; Encoding the electric energy influencing factor information in the multi-dimensional data set to generate external influencing features; Combining the behavior features and the external influencing features to form multi-dimensional initial features; Dimension reduction is performed on the multi-dimensional initial features to obtain the core feature set. 2.The AI model-based control method of a light storage and charging energy source unit according to claim 1, wherein, The step of controlling the light storage and charging energy unit using an AI control model corresponding to the unit grouping comprises: Obtaining a time series prediction network model and the AI control model pre-constructed for each unit grouping; Using the time series prediction network model to predict real-time data of the light storage and charging energy unit to obtain electric energy prediction data; Generating control instructions from the AI control model according to the real-time data of the light storage and charging energy unit and the electric energy prediction data, and executing the control instructions by the light storage and charging energy unit. 3.The AI model-based control method of a light storage and charging energy source unit according to claim 2, wherein, The step of predicting real-time data of the light storage and charging energy unit using the time series prediction network model comprises: Collecting real-time data of the light storage and charging energy unit, including real-time electric energy data and real-time electric energy influencing factor information; Determining the unit grouping to which the light storage and charging energy unit belongs; Using the time series prediction network model of the unit grouping to which the light storage and charging energy unit belongs to predict the real-time electric energy data and the real-time electric energy influencing factor information to obtain the electric energy prediction data.

4. The AI model-based control method of the light storage and charging energy unit according to claim 1, wherein The step of collecting electric energy historical data of the light storage and charging energy unit comprises: collecting original electric energy data from intelligent electric meter terminals and power supply and consumption terminals of the light storage and charging energy unit, and performing numerical processing on the original electric energy data according to a preset sampling period to obtain the electric energy historical data; The step of obtaining electric energy influencing factor information comprises: obtaining meteorological data from an environmental data management platform, and determining date marker data corresponding to the meteorological data according to a calendar database, and performing normalization processing on the meteorological data and the date marker data to obtain the electric energy influencing factor information. 5.The AI model-based control method of a light storage and charging energy source unit according to claim 1, wherein, The step of processing the electric energy historical data and the electric energy influencing factor information into a multi-dimensional data set with unified data format comprises: Performing time alignment processing on the electric energy historical data and the electric energy influencing factor information to ensure that they correspond in the time dimension. The power history data and the power influencing factor information after the time alignment processing are converted into a preset data format; The power history data and the power influencing factor information after the data format conversion are arranged into the multi-dimensional data set. 6.The AI model-based control method of a light storage and charging energy source unit according to claim 1, wherein, The step of reducing the dimensionality of the multi-dimensional initial features includes: calculating the covariance matrix of the multi-dimensional initial features; solving the eigenvalues and corresponding orthogonal unit eigenvectors of the covariance matrix; sorting the dimensions of the multi-dimensional initial features according to the eigenvalues; extracting the features corresponding to the top-ranked eigenvalues to form the multi-dimensional initial features after dimensionality reduction. 7.The AI model-based control method of a light storage and charging energy source unit according to claim 6, wherein, The step of extracting the features corresponding to the top-ranked eigenvalues includes: calculating the cumulative variance contribution rate in turn according to the order of the eigenvalues; determining the dimensionality of the extracted features according to the size of the cumulative variance contribution rate. 8.The AI model-based control method of a light storage and charging energy source unit according to any one of claims 1 to 7, wherein, The step of using the core feature set to cluster the optical storage and charging energy units includes: calculating the similarity between the core feature sets corresponding to the optical storage and charging energy units; merging the optical storage and charging energy units into multiple initial unit groups using a hierarchical clustering method according to the similarity; optimizing the multiple initial unit groups using a fuzzy mean clustering optimized by a genetic algorithm to obtain unit groups with similar power application behaviors.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the control method of the optical storage and charging energy unit based on the AI model according to any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the control method of the optical storage and charging energy unit based on the AI model according to any one of claims 1 to 8.

11. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the processor to perform the method of any one of claims 1-10. The processor executes the computer program to implement the steps of the control method of the optical storage and charging energy unit based on the AI model according to any one of claims 1 to 8.

Images