A cross-cycle multi-source heterogeneous power data processing system and application method

Abstract

A cross-cycle multi-source heterogeneous power data processing system includes a data acquisition module, a data analysis module, a data modeling module, a multi-source heterogeneous information module, a prediction module, a switching module, and a data push module. These modules are application software installed on a PC. An application method for the cross-cycle multi-source heterogeneous power data processing system includes seven steps. This invention can derive a power load data model for a power consumption area and predict power consumption based on the obtained model. It can adjust power consumption in advance before changes occur in the area's power consumption, ensuring stable power supply on-site. It can also promptly alert both the power supply and consumption ends when power consumption data is abnormal, reducing the probability of power accidents. Furthermore, it can provide users with reasonable energy-saving suggestions and proactively prompt users to conduct maintenance when there are significant deviations between the power supply data and normal operating conditions.

Description

Technical Field

[0001] This invention relates to the field of power technology, and in particular to a multi-source heterogeneous power data processing system and application method that spans multiple cycles. Background Technology

[0002] With the development of power supply technology, in order to ensure safe power supply, power departments or users of relevant power terminals will install monitoring equipment such as temperature, humidity, voltage and current in the corresponding areas in addition to electricity meters, to monitor the working performance of electrical equipment in the corresponding areas and whether the power supply is normal, so as to ensure safe power supply as much as possible.

[0003] While existing power monitoring equipment has achieved intelligent power supply management to some extent, it still has certain shortcomings due to structural limitations, specifically as follows: First, the monitored data can only reflect the power consumption situation at that time. Both the power user and the power supplier can only adjust their power consumption based on the current data. For example, if a peak in power consumption occurs in a certain area during a certain period, leading to a drop in power supply voltage, the power supplier can allocate sufficient power to meet the operating needs of the equipment in that area, or the power user can reduce some load at that moment to ensure the normal operation of important equipment. However, this passive control method means that neither the power user nor the power supplier can receive advance warnings, which hinders timely response and negatively impacts power supply safety. Second, it cannot alert the power user when a potential power outage (such as a short circuit or open circuit) is expected. Typically, the power user only learns of the situation after observing abnormal data from the monitoring equipment. This prevents the development of targeted contingency plans in advance, also negatively impacting power supply safety. In conclusion, it is particularly necessary to provide a system and application method that can analyze and model users' daily electricity consumption behavior and ensure safe and stable power supply to users as much as possible. Summary of the Invention

[0004] To overcome the shortcomings of existing power supply technologies, as described in the background, due to technological limitations, this invention provides a multi-source heterogeneous power data processing system and application method that integrates data collected by various sensors and, through the joint action of corresponding module units, performs unified calculation, analysis, and modeling of the data. The modeled data can automatically and adaptively adjust the power load of relevant areas at different time periods to ensure stable power supply on site. Furthermore, it can provide users with reasonable energy-saving suggestions based on the obtained data and proactively prompt the power user to carry out maintenance when there is a significant deviation between the power supply data and normal conditions (indicating a probability of failure). This effectively ensures safe power supply.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A cross-cycle multi-source heterogeneous power data processing system, characterized by comprising a data acquisition module, a data analysis module, a data modeling module, a multi-source heterogeneous information module, a prediction module, a switching module, and a data push module; wherein the data acquisition module, data analysis module, data modeling module, multi-source heterogeneous information module, prediction module, switching module, and data push module are application software installed on a PC; an application method of the cross-cycle multi-source heterogeneous power data processing system includes the following steps: Step A: The data acquisition module collects power consumption area data, performs preliminary classification processing on the above data, and outputs it to the data analysis module; Step B: The data analysis module, based on artificial intelligence technology, uses one or more of weakly supervised / semi-supervised / unsupervised machine learning methods to carry out a data-driven user classification algorithm, calculates and analyzes the input data obtained by the data acquisition module, and constructs a user classification algorithm model based on time-series behavior patterns of users' power consumption. Specific user data used includes 96-point curves, daily power consumption, and monthly power consumption data; Step C: The data modeling module performs characteristic attribution analysis on the data classification algorithm results obtained in Step B, specifically, based on the weakly supervised learning classification algorithm model... The classification results are analyzed for attribution and visualization to extract relevant characteristics of user electricity consumption, and the user classification results are used to serve electricity consumption and load forecasting. Step D: The multi-source heterogeneous information module constructs a cross-cycle accurate electricity load forecasting model based on the data from Step C and the multi-source heterogeneous information. Specifically, the multi-source heterogeneous information includes not only the 96-point curve, daily and monthly electricity consumption data, but also factors such as meteorology, economy, policy, and electricity consumption change processes. Change processes include capacity increases / decreases, classification changes, and transfers of ownership in electricity consumption areas. The data obtained in this step serves user electricity consumption forecasting. Step E: The forecasting module is based on multi-source... Heterogeneous feature representation is used to calculate and establish a time-series learning model for accurate load forecasting across cycles. Specifically, the model data includes 96-point curves (30 days * 96) predicting the entire month of electricity consumption area one year, one quarter, and one month in advance, as well as 96-point curves predicting T+2 days from day T (before 14:00). Step F: When the load of users in the electricity consumption area is abnormal, the data push module pushes prompt data to the electricity user and the management, and gives the specific abnormality type. Step G: The switching module adjusts the power supply power of the power supply terminal to the power supply area based on the electricity consumption forecast data of the historical data of the electricity consumption area during the peak or off-peak periods of the electricity consumption area.

[0007] Furthermore, in step A, the data acquisition module collects data from the electricity consumption area, including electricity consumption data per unit time obtained from meters, ammeters, voltmeters, and thermometers, current and voltage data for each time period, and on-site temperature data.

[0008] Furthermore, in step C, the electricity and load prediction data obtained by the user classification can be displayed through the display interface and output to the switching module.

[0009] Furthermore, in step F, the data push module determines that the abnormal load data of users in the power consumption area includes abnormally high current, abnormally low voltage, and excessively high temperature. Specifically, the database submodule of the data push module stores historical average current, voltage, and temperature data of the corresponding power consumption area for the corresponding time period. When the above data exceeds the threshold, information is pushed.

[0010] Furthermore, in step G, while the switching module adjusts the power supply power of the power supply terminal to the power supply area during peak or off-peak periods, it will also adjust the output power in a targeted manner based on the feedback data of the relevant areas to meet the power needs of the on-site electrical equipment.

[0011] Furthermore, the data push module can also provide reasonable energy-saving suggestions based on electricity consumption data, thereby achieving the goal of energy saving and emission reduction.

[0012] Furthermore, step E includes determining a unified feature representation method for multi-source heterogeneous information and determining a time-series learning model for cross-cycle accurate load forecasting. The determination of the unified feature representation method for multi-source heterogeneous information includes constructing a basic model using the random forest algorithm and completing the fusion of multi-source heterogeneous data based on model ensemble. The determination of the time-series learning model for cross-cycle accurate load forecasting is a cross-cycle accurate load forecasting model established on the feature representation of multi-source heterogeneous information, which mainly uses a grey prediction model to complete load forecasting.

[0013] The beneficial effects of this invention are as follows: Based on data collected from multiple sensors, and through the combined action of related software units, this invention can uniformly analyze and model the data to derive a power load data model for a given area within a given time period. It can also predict power consumption based on this model, adjusting the power supply in advance before changes in power consumption occur in the area, ensuring stable power supply on-site. Furthermore, it can promptly alert both the power supply and user ends when power consumption data is abnormal, indicating a possibility of short circuits or open circuits, thus reducing the probability of power accidents. This invention can also provide users with reasonable energy-saving suggestions based on the acquired data, and proactively alert users to conduct maintenance when there are significant deviations between the power supply data and normal conditions, effectively ensuring safe power supply. Based on the above, this invention has promising application prospects. Attached Figure Description

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Figure 1 This is a block diagram of the architecture of the present invention.

[0016] Figure 2This is a flowchart of the workflow architecture of the K-means clustering algorithm of this invention.

[0017] Figure 3 This is a block diagram of the semi-supervised SSCADP algorithm architecture of the present invention. Detailed Implementation

[0018] Figure 1 As shown, a multi-cycle multi-source heterogeneous power data processing system includes a data acquisition module, a data analysis module, a data modeling module, a multi-source heterogeneous information module, a prediction module, a switching module, and a data push module; the data acquisition module, data analysis module, data modeling module, multi-source heterogeneous information module, prediction module, switching module, and data push module are application software installed in the PC at the power supply end.

[0019] Figure 1 As shown, an application method for a multi-source heterogeneous power data processing system across cycles includes the following steps: Step A: The data acquisition module collects power consumption data of the power consumption area in real time, performs preliminary classification and processing of the above data, and outputs it to the data analysis module to provide basic calculation data for the data analysis module; Specifically, in Step A, the data acquisition module collects data of the power consumption area, including total power consumption data per unit time, current and voltage data for each time period, and specific temperature data obtained from meters, ammeters, voltmeters, thermometers, etc. The data acquisition module can store the above data for easy retrieval later.

[0020] Figure 1As shown, step B: The data analysis module, based on artificial intelligence, employs one or more methods of weakly supervised / semi-supervised / unsupervised learning to conduct a data-driven user classification algorithm. It calculates and analyzes the obtained total electricity consumption data, current and voltage data for each time period, and specific on-site temperature data to derive user-based electricity consumption time-series behavior patterns, which are then used to construct a user classification algorithm model based on time-series behavior. Specific user data used includes a 96-point curve (a curve established by sampling data at 15-minute intervals), daily electricity consumption, and monthly electricity consumption data. The K-means algorithm (unsupervised learning) specifically used in this invention is one of the most widely used clustering algorithms. Each cluster is represented by the average value of all resources within it; this average value is the centroid of each cluster in the clustering result. The core idea of ​​the K-Means algorithm is to randomly select K objects (total electricity consumption data, current and voltage data for each time period, and specific on-site temperature data, etc.) as centroids in a certain space, classify data objects close to the centroids, and then recalculate the centroids. This process is iteratively updated until the algorithm's rule function converges to an optimal state. The K-means algorithm uses the sum of squared errors (SSE) as a quality metric. When two different clustering results are obtained (different total electricity consumption data, current and voltage data for different time periods, and specific on-site temperature data, etc.), the result with the lower SSE is given priority. The formula for calculating SSE is: The formula for calculating the cluster centroid ei is: In the formula, K is the number of generated clusters, Ei is the i-th cluster, ei represents the centroid of cluster Ei, ni represents the number of data objects contained in the i-th cluster, and x represents a data object. 1) The clustering algorithm implementation process is as follows: ① Input: The calculation basis is the dataset Ω, n data objects (different total power consumption data, current and voltage data for each time period, and specific temperature data on site, etc.), the formula is as follows:

[0021] Ω={x i |x i =(x i1 x i2 , ..., x id ), i = 1, 2, ..., n}

[0022] C = {c j |c j =(c j =(c j1 c j2 c jd ), j = 1, 2, ..., k)}

[0023] In the formula, x i =(x i1 x i2, ..., x id ) represents a d-dimensional data vector, representing all (d) attributes of any given data, c j =(c j =(c j1 c j2 c jd ) represents the centroid of the j-th cluster, n is the size of the set; K is the number of clusters, and C represents the set of all cluster centroids. ② Output: Obtain the clustering result R, the formula is:

[0024] R = {r j =(r j1 r j2 ,...r jh ), j = 1, 2...k, 0 < k < n}. 2) Specific steps: ① Set any two data objects (different total power consumption data, current and voltage data for each time period, and specific temperature data on site, etc.) The Euclidean distance between ci and cj is dis(c i c j Specifically, it is expressed as:

[0025] Calculate the distance from each data object in the dataset Ω to the K cluster centroids, and merge the data objects into the clusters with the closest distance. ② The cluster centroid cj is represented as: Where N(ψi) is the number of data points in the same cluster; recalculate the centroid of each cluster. ③ Determine if the termination condition has been met; if so, output the result; if not, jump back to step ② to continue the calculation; the K-means clustering algorithm workflow is as follows: Figure 2 As shown. The semi-supervised SSCADP algorithm framework used in this invention is as follows. Figure 3As shown, the SSCAPP algorithm uses a Hoeffding tree as the base classifier. After each sample in tD is classified, each sample, starting from the root node, falls sequentially into the leaf nodes of the decision tree according to the splitting attribute and splitting value. The statistical values ​​corresponding to that leaf node (different total electricity consumption data, current and voltage data for different time periods, and specific temperature data on site, etc.) will be updated, such as the total number of samples, the values ​​of each attribute, and the distribution of categories. Once the current decision tree has satisfied a detection cycle, a density-adaptive clustering algorithm is called in each leaf node to label unlabeled samples using a small amount of sample labeling information, and concept drift detection is performed simultaneously. Then, the detected local structures with concept drift are pruned to adapt to the current concept. Finally, if the number of samples in the leaf node reaches a specified threshold, the Hoeffding inequality is used to determine whether the current node satisfies the splitting condition. SSCADP uses a Hoeffding tree as its base classifier. When a training sample arrives, it recursively falls into the corresponding leaf node along the root node based on the splitting attributes and values ​​of each node. Simultaneously, the information of this sample point is stored to update the statistical values ​​of the current leaf node. Subsequently, once a detection cycle is met, unlabeled samples are labeled using a density-adaptive clustering method. When the number of samples in a leaf node (different total electricity consumption data, current and voltage data for different time periods, and specific on-site temperature data, etc.) reaches a specified threshold, a splitting mechanism is initiated based on the Hoeffding inequality and previously recorded statistical information. If a detection cycle is reached, SSCADP calls an adaptive cluster center localization algorithm, which automatically locates cluster centers by combining the change detection method in SAND with the CFSDP algorithm. After concept clusters are formed in each leaf node, a graph-based label propagation algorithm uses a small number of labeled samples in each cluster to label unlabeled samples. If a cluster consists entirely of unlabeled samples, all samples are labeled with the majority class of the nearest cluster.

[0026] Figure 1 As shown in the figure, step C: The data modeling module performs characteristic attribution analysis on the data classification algorithm results obtained in step B. Specifically, based on the classification algorithm model of weakly supervised learning, it conducts attribution analysis and visualization analysis on the classification results, extracts the characteristics of user electricity consumption (such as the electricity load of each time period), and uses the above user classification results to predict electricity consumption and load.

[0027] Figure 1As shown in step C, the electricity consumption and load forecast data obtained by the user classification can be displayed on the display interface and stored, allowing staff to access the data at any time. The data is then output to the switching module. Step D: The multi-source heterogeneous information module constructs a cross-cycle accurate electricity load forecasting model based on the data from step C and combined with multi-source heterogeneous information. Specifically, the multi-source heterogeneous information also includes meteorological data, economic and policy impacts, and electricity consumption change processes (user capacity increases / decreases, classification changes, transfers, etc.). Due to the more diverse input data, it can effectively serve the user's electricity consumption forecasting needs.

[0028] Figure 1As shown in the figure, step E: The prediction module calculates and establishes a time series learning model for accurate user load prediction across cycles based on the multi-source heterogeneous feature representation. Specifically, the time series learning model includes establishing a 96-point curve (30 days * 96) for predicting the entire month of electricity consumption area one year, one quarter, and one month in advance, as well as a 96-point curve for predicting T+2 days on day T (before 14:00). Step F: When the load of users in the power consumption area is abnormal, the data push module pushes prompt data (voice, SMS, text, etc.) to the smartphones and PCs of the power users and the management, and gives the specific abnormality type (for example, the current is too high on site, which may be due to a short circuit of the power equipment; the current is too low, which may be due to an open circuit of the power supply). The data push module judges the abnormal load of users in the power consumption area, including abnormally high current, abnormally low voltage, and excessively high temperature. Specifically, the database sub-module of the data push module stores the historical average current, voltage, and temperature data of the corresponding power consumption area for the corresponding time period. The data push module can compare the real-time collected data with the data of the database sub-module. When the above data exceeds the threshold, the information is pushed. The data push module can also give reasonable energy-saving suggestions based on the power consumption data to achieve the purpose of energy saving and emission reduction. This invention conducts attribution analysis and visualization analysis of the classification results based on the weakly supervised learning classification algorithm model, and extracts the characteristics of user power consumption. The results of user classification are used to predict power consumption and load, specifically including the following steps. Research on unified feature representation method of multi-source heterogeneous information, (1) Use random forest algorithm to build basic model. The random forest algorithm randomly extracts data attribute values ​​for model construction in different grids, which can effectively solve the attribute differences of multi-source heterogeneous data and complete model training. Secondly, for a large amount of unlabeled data, the semi-supervised random forest model is modified by pre-pruning. That is, the basic random forest model built based on labeled data is validated after adding unlabeled data. If the new model improves the effect, the pruning is retained; otherwise, it is rolled back. The pruning is optimized and solved by the out-of-bag estimation algorithm. (2) Multi-source heterogeneous data fusion is completed based on model ensemble. The temporal attributes, spatial topological attributes and other features of multi-source heterogeneous information are analyzed and modeled. Sub-classifiers are built for the features of each set of structured data. Then, the multiple sub-classifiers are integrated into the model to realize multi-source heterogeneous data fusion based on model ensemble. This research focuses on time-series learning models for accurate load forecasting across cycles. Specifically, it investigates a model based on multi-source heterogeneous information feature representation, employing a grey forecasting model to complete load prediction. Grey system theory posits that all random variations, i.e., grey metrics, change within a certain range. Typically, methods of cumulative generation and cumulative subtraction are used to transform chaotic, unorganized raw data into a regular sequence of generated data. Because the grey forecasting model is based on the generated data, the grey forecasting data requires restoration processing to the predicted values ​​obtained from the generated data model.When the differential equation of the grey model is used as a single indicator of the power system (such as load), the expression of the time response function of the solution of the differential equation is the grey prediction model. After verifying and correcting the accuracy and reliability of the model, this model can be used to predict the future development and change of power load.

[0029] Figure 1 As shown in step G: The switching module adjusts the power supply power of the power supply terminal to the power supply area in advance based on the power consumption forecast data of the historical data of the power consumption area during the peak or off-peak periods of the power consumption area. Specifically, in step G, while adjusting the power supply power of the power supply terminal to the power supply area during the peak or off-peak periods of the power consumption area, the switching module will also adjust the output power in a targeted manner based on the feedback data of the relevant area to meet the power consumption needs of the on-site electrical equipment.

[0030] Figure 1 As shown, this invention, based on data collected from multiple sensors, utilizes a data acquisition module, a data analysis module, a data modeling module, a multi-source heterogeneous information module, a prediction module, a switching module, and a data push module to uniformly analyze and model the data, deriving a power load data model for a given area per unit time. Based on this model, it can predict power consumption and adjust power usage in advance before changes occur, ensuring stable power supply. Furthermore, it can promptly alert both the power supply and user ends when power data is abnormal, indicating a possibility of short circuits or open circuits, reducing the likelihood of power accidents. This invention can also provide users with reasonable energy-saving suggestions based on the acquired data and proactively alert users to conduct maintenance when power supply data deviates significantly from normal levels, effectively ensuring safe power supply.

[0031] The foregoing has shown and described the basic principles and main features of the present invention, as well as its advantages. It will be apparent to those skilled in the art that the present invention is limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or basic characteristics. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

[0032] Furthermore, it should be understood that although this specification describes the embodiments, the embodiments do not necessarily contain only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An application method for a multi-source heterogeneous power data processing system spanning multiple cycles, characterized in that, The system includes a data acquisition module, a data analysis module, a data modeling module, a multi-source heterogeneous information module, a prediction module, a switching module, and a data push module. These modules are application software installed on a PC. The system comprises the following steps: Step A: The data acquisition module collects electricity consumption area data, performs preliminary classification processing, and outputs the data to the data analysis module. Step B: The data analysis module, based on artificial intelligence technology, uses one or more of weakly supervised / semi-supervised / unsupervised machine learning methods to conduct a data-driven user classification algorithm and analyze the input data obtained from the data acquisition module. Based on calculations and analysis, and considering users' electricity consumption time-series behavior patterns, a user classification algorithm model based on time-series behavior is constructed. Specific user data used includes 96-point curves, daily electricity consumption, and monthly electricity consumption data. Step C: The data modeling module performs characteristic attribution analysis on the data classification algorithm results obtained in Step B. Specifically, based on the weakly supervised learning classification algorithm model, attribution analysis and visualization analysis are conducted on the classification results to extract relevant characteristics of user electricity consumption, and the user classification results are used to serve electricity consumption and load forecasting. Step D: The multi-source heterogeneous information module constructs a cross-cycle accurate electricity load forecasting model using the data from Step C and multi-source heterogeneous information. Specifically, multi-source heterogeneous information... In addition to the 96-point curve, daily and monthly electricity consumption data, the data also includes meteorological, economic, policy, and electricity consumption change process factors. The change process includes capacity increases / decreases, category changes, and transfers of ownership for electricity consumption areas. The data obtained in this step serves user electricity consumption forecasting. Step E: The forecasting module, based on multi-source heterogeneous feature representation, calculates and establishes a time-series learning model for accurate load forecasting across cycles. Specifically, the model data includes the 96-point curves for the entire month of electricity consumption areas predicted one year, one quarter, and one month in advance, as well as the 96-point curves for day T predicting day T+2. Step F: The data push module pushes alert data to electricity users and management when user load in the electricity consumption area is abnormal, and provides the specific type of abnormality. Step G: The switching module, based on the electricity consumption area... Historical electricity consumption forecast data is used to adjust the power supply power of the power supply terminal to the power supply area during peak or off-peak periods. In step F, the data push module judges abnormal load data of users in the power consumption area, including abnormally high current, abnormally low voltage, and excessively high temperature. Specifically, the database submodule of the data push module stores historical average current, voltage, and temperature data of the corresponding power consumption area for the corresponding time period. When the above data exceeds the threshold, information is pushed. In step G, while the switching module adjusts the power supply power of the power supply terminal to the power supply area during peak or off-peak periods, it will also adjust the output power in a targeted manner based on the feedback data of the relevant area to meet the power needs of the on-site electrical equipment.

2. The application method of a multi-source heterogeneous power data processing system across cycles according to claim 1, characterized in that, In step A, the data acquisition module collects data from the electricity consumption area, including electricity consumption data per unit time obtained from meters, ammeters, voltmeters, and thermometers, current and voltage data for each time period, and on-site temperature data.

3. The application method of a multi-source heterogeneous power data processing system across cycles according to claim 1, characterized in that, In step C, the electricity and load forecast data obtained by the user classification can be displayed on the display interface and output to the switching module.

4. The application method of a multi-source heterogeneous power data processing system across cycles according to claim 1, characterized in that, The data push module can also provide reasonable energy-saving suggestions based on electricity consumption data, so as to achieve the goal of energy saving and emission reduction.

5. The application method of a multi-source heterogeneous power data processing system across cycles according to claim 1, characterized in that, Step E includes determining a unified feature representation method for multi-source heterogeneous information and determining a time-series learning model for cross-cycle accurate load forecasting. The determination of the unified feature representation method for multi-source heterogeneous information includes constructing a basic model using the random forest algorithm and completing the fusion of multi-source heterogeneous data based on model ensemble. The determination of the time-series learning model for cross-cycle accurate load forecasting is a cross-cycle accurate load forecasting model established on the feature representation of multi-source heterogeneous information, which mainly uses the grey prediction model to complete load forecasting.

Images