Power grid operation safety analysis method based on multi-source data fusion and big data

Abstract

The present application relates to the technical field of power system, and especially relates to a power grid operation safety analysis method based on multi-source data fusion and big data. The power grid operation safety analysis method based on multi-source data fusion and big data comprises the following steps: S1: collecting power distribution network topology structure data and power distribution network region data, dividing basic power supply units based on power distribution network topology nodes based on the power distribution network topology structure data and the power distribution network region data, and generating a power distribution network power supply diagram; S2: clustering behavior patterns according to the power distribution network power supply diagram and historical mobile signaling data to obtain user group distribution of each basic power supply unit; and S3: analyzing historical power consumption data according to weather, time period and date type to obtain basic power consumption parameters of user groups. The present application can generate early warning results with clear positioning information by combining the power consumption habits of users, and prevent overload or fault diffusion.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method for power grid operation safety analysis based on multi-source data fusion and big data. Background Technology

[0002] With the accelerating pace of urbanization and the continuous expansion of the power system, the distribution network, as a crucial link directly facing end users, is receiving increasing attention for its operational status and power supply reliability. Especially in complex urban areas, distribution networks often exhibit characteristics such as complex topologies, dense power supply units, diverse user types, and significant load fluctuations. Electricity consumption behaviors vary significantly across different regions at different times, under different weather conditions, and on different dates, resulting in a highly dynamic and uncertain operational status for the distribution network. As residents' lifestyles, business activities, and the operational modes of public service facilities continue to change, traditional experience-based or static statistical methods for distribution network operation analysis and load forecasting are no longer sufficient to accurately reflect the actual changes in distribution network operation. Therefore, there is an urgent need for a power grid operation safety analysis method based on multi-source data fusion and big data, capable of generating early warning results with clear location information by combining user electricity consumption habits. This would enable maintenance personnel to develop intervention measures in advance for specific power supply units or topological weaknesses, effectively preventing overloads or the spread of faults. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies that ignore the impact of user groups on the power distribution network, this invention provides a power grid operation safety analysis method based on multi-source data fusion and big data.

[0004] The technical implementation scheme of the present invention is as follows: a power grid operation safety analysis method based on multi-source data fusion and big data, comprising the following steps:

[0005] S1: Collect distribution network topology data and distribution network area data, and based on the distribution network topology data and distribution network area data, divide the basic power supply units according to the distribution network topology nodes to generate a distribution network power supply diagram;

[0006] S2: Based on the power distribution network diagram and historical mobile signaling data, perform behavioral pattern clustering to obtain the user group distribution of each basic power supply unit;

[0007] S3: Based on historical electricity consumption data, analyze the data according to weather, time period, and date type to obtain basic electricity consumption parameters for user groups;

[0008] S4: Obtain meteorological data for future periods, combine user group distribution and basic electricity consumption parameters to predict the electricity load of each basic power supply unit and generate a distribution network load forecast map;

[0009] S5: Perform operational safety analysis based on the power distribution network load forecast diagram to generate early warning information, and assist operation and maintenance personnel in making countermeasure decisions based on the early warning information.

[0010] Preferably, the step of collecting distribution network topology data and distribution network area data, and dividing basic power supply units based on distribution network topology nodes to generate a distribution network power supply map includes: obtaining distribution network topology data and distribution network area data from a distribution network management database, wherein the distribution network area data includes area function type, infrastructure and area boundary; dividing basic power supply units based on distribution topology nodes in the distribution network topology data and distribution area data; and generating a distribution network power supply map based on the divided N basic power supply units.

[0011] Preferably, behavioral pattern clustering is performed based on the power distribution network diagram and historical mobile signaling data to obtain the user group distribution of each basic power supply unit. This includes: acquiring anonymized historical mobile signaling data, which includes historical mobile signaling data for weekdays, holidays, and days with school closures and work-from-home arrangements; performing hierarchical clustering based on dynamic time warping on the historical mobile signaling data based on date type and time period to generate user group clusters, which include stable residential, regular office, public medical, educational, and commercial / entertainment types; statistically analyzing user group clusters based on date type and time period to obtain the total number of user group baselines and constructing a user group conversion probability matrix; calculating the total number of user groups based on the total number of user group baselines and the user group conversion probability matrix; and fitting the distribution of user group numbers within each basic power supply unit using a conditional Gaussian mixture model based on the power distribution network diagram.

[0012] Preferably, the step of obtaining the total number of user group baselines and constructing a user group conversion probability matrix by statistically analyzing user group clusters based on date type and time period includes: constructing a continuous-time Markov transition intensity matrix with weather, time period and date type as condition variables, statistically quantifying the conversion probability between different user groups, and training the parameters in the transition intensity matrix through maximum likelihood estimation to obtain the user group conversion probability matrix.

[0013] Preferably, the distribution of the number of users in each basic power supply unit is fitted using a conditional Gaussian mixture model based on the power distribution network diagram. This includes: fitting the distribution of the number of users in each basic power supply unit using a conditional Gaussian mixture model based on the historical mobile signaling data and user group clusters; constructing conditional variables according to weather, time period, and date types to train the conditional Gaussian mixture model according to the user group clusters; obtaining the distribution of the number of users in each basic power supply unit under different conditional variables; and calculating the number of users in each basic power supply unit based on the conditional variables and the distribution of the number of users.

[0014] Preferably, calculating the number of users within each basic power supply unit based on the condition variable and the user group size distribution includes: the user group size distribution formula is:

[0015] ;

[0016] In the formula, Basic power supply unit Inner Number of users in this category For the first Total number of users in this category Basic power supply unit , For condition variables Next User groups in basic power supply units The probability density function of occurrence.

[0017] Preferably, the step of analyzing historical electricity consumption data according to weather, time period, and date types to obtain basic electricity consumption parameters for user groups includes: acquiring historical electricity consumption data for different user groups and electricity consumption data for infrastructure in different basic power supply units; constructing conditional variables based on weather, time period, and date types to perform piecewise linear regression analysis on the historical electricity consumption data; constructing electricity load curves for different user groups and infrastructure; and obtaining the basic electricity load of individual users for different user groups and the basic electricity load of infrastructure in different power supply units based on the electricity load curves.

[0018] Preferably, the step of acquiring meteorological data for future periods, combining user group distribution and basic electricity consumption parameters to predict the electricity load of each basic power supply unit and generate a distribution network load prediction map includes: constructing future conditional variables based on date type and meteorological data for future periods, calculating the electricity load of each basic power supply unit based on the future conditional variables, and generating a distribution network load prediction map based on the distribution network power supply map and the electricity load of each basic power supply unit.

[0019] Preferably, calculating the power load of each basic power supply unit based on the future condition variables includes: calculating the power load of each basic power supply unit using a load forecasting calculation formula based on the number of user groups within each basic power supply unit, the basic power load per user of different user groups, and the basic power load of infrastructure in different power supply units. The load forecasting calculation formula is as follows:

[0020] ;

[0021] In the formula, For the electrical load of the basic power supply unit, For condition variables Lower user groups The basic electricity load of a single user The electrical load of the infrastructure for the basic power supply unit.

[0022] Preferably, the operation safety analysis based on the distribution network load forecast map generates early warning information, and the early warning information assists operation and maintenance personnel in making countermeasure decisions, including: performing operation safety analysis based on the distribution network load forecast map using a graph neural network; defining a graph structure based on the distribution network load forecast map, with basic power supply units as nodes and the connection relationships of basic power supply units as edges; constructing node feature vectors based on meteorological data in future time periods, the power load of each basic power supply source, and the number of user groups in each basic power supply unit; performing safety operation analysis using a graph attention network; generating early warning information containing basic power supply units and topology; and assisting operation and maintenance personnel in making countermeasure decisions for specific basic power supply units and topology based on the early warning information.

[0023] The beneficial effects are as follows: This invention integrates the distribution network topology, regional functional data and user mobile signaling behavior patterns, divides the data at the basic power supply unit level, combines a conditional Gaussian mixture model to quantify the dynamic distribution of user groups, and integrates meteorological and time period data to construct multi-dimensional conditional variables, thereby improving the spatiotemporal accuracy of load forecasting and overcoming the shortcomings of traditional methods that rely on historical electricity consumption data and ignore the spatial heterogeneity of behavior.

[0024] This invention performs in-depth analysis of the load prediction map of the distribution network based on graph neural network. It constructs a graph structure with basic power supply units as nodes and connection relationships as edges, and combines node feature vectors to realize intelligent assessment of operational risks, generating early warning results with clear location information. This enables operation and maintenance personnel to formulate intervention measures in advance for specific power supply units or topological weak points, effectively preventing overload or fault spread.

[0025] This invention employs dynamic time-warping hierarchical clustering to identify diverse user behavior patterns, and utilizes conditional variables to drive Gaussian mixture models and piecewise linear regression, enabling the prediction model to flexibly adapt to complex scenarios such as weather changes and holidays, thereby enhancing the response speed to sudden load fluctuations and the stability of long-term predictions. Attached Figure Description

[0026] Figure 1 This is a flowchart of the power grid operation safety analysis method of the present invention;

[0027] Figure 2 This is a diagram illustrating the power grid operation safety analysis and decision-making process of the present invention. Detailed Implementation

[0028] The invention will now be described more fully below with reference to the accompanying drawings, in which presently preferred embodiments of the invention are illustrated. However, the invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided for thoroughness and completeness and to fully convey the scope of the invention to those skilled in the art.

[0029] Power grid operation safety analysis methods based on multi-source data fusion and big data, such as Figure 1 and Figure 2 As shown, it includes the following steps:

[0030] S1: Collect distribution network topology data and distribution network area data, and based on the distribution network topology data and distribution network area data, divide the basic power supply units according to the distribution network topology nodes to generate a distribution network power supply diagram;

[0031] The distribution network topology data and distribution network area data are obtained from the distribution network management database. The distribution network area data includes area function type, infrastructure and area boundary. Based on the distribution topology nodes in the distribution network topology data and the distribution area data, power supply accessibility analysis is performed to divide the distribution into basic power supply units. A distribution network power supply map is generated based on the divided N basic power supply units.

[0032] It needs further explanation that the current distribution network topology data and distribution network area data are retrieved from the distribution network management database. Distribution network topology data describes the connection relationships between various distribution topology nodes in the distribution network, including substation nodes, feeder nodes, switch nodes, and the line connections between these nodes, thus comprehensively reflecting the physical structure and electrical connections of the distribution network. Distribution network area data describes the geographical attributes covered by the distribution network, including the corresponding regional function type, infrastructure within the region, and regional boundary information. The regional function type characterizes the main usage attributes of the region, such as schools and hospitals; infrastructure characterizes the distribution of fixed electricity users within the region; and regional boundaries define the spatial scope of the region. After obtaining the distribution network topology data and distribution network area data, a power supply accessibility analysis is performed on the distribution network based on the distribution network topology nodes and the distribution network area data. Specifically, firstly, a distribution topology node in the distribution network topology is selected as the starting point for the analysis. This node typically represents a source with power supply capability, such as an outgoing switch of a substation or an important distribution ring main unit. Based on the electrical connection paths defined in the topology data, starting from the source point, the process traverses downstream along all energized lines, logically outlining all electrical paths from the source point through the network that can deliver power. While traversing these electrical paths, the analysis process integrates distribution area data. For each geographical area traversed or covered by an electrical path, the area's functional type and infrastructure information are retrieved. The regional functional type influences the prediction of load characteristics; for example, the load curve of a commercial center differs from that of a residential area, requiring differentiation. Infrastructure information directly contributes to the guaranteed basic load. Regional boundary data provides spatial constraints. Power supply accessibility analysis determines the geographical boundaries that the power supply can stably cover from the current source point, while meeting necessary voltage quality and line capacity constraints. Through this power supply accessibility analysis, a continuous geographical range stably powered by the same or a group of electrically closely related distribution topology nodes can be clearly defined; this range is defined as a basic power supply unit. Each basic power supply unit has a clearly defined electrical power source, a clear and continuous spatial boundary, and relatively consistent attributes in terms of social function. Depending on the complexity of the actual power grid, this process can divide the system into N basic power supply units, where N is a positive integer.

[0033] Finally, a distribution network power supply diagram is generated based on the N basic power supply units. This diagram represents a new type of digital representation of the power grid. In the distribution network power supply diagram, each basic power supply unit is represented as an independent graphical block, with information on its corresponding key distribution topology nodes, a summary of the regional functional type, and a list of important infrastructures labeled within the block. Simultaneously, the diagram clearly illustrates the electrical connections between the basic power supply units formed through the power grid topology using lines.

[0034] Through the above steps, using the distribution network topology data and distribution network area data obtained from the distribution network management database, based on the distribution topology nodes, and combined with the regional functional type, infrastructure and regional boundaries, power supply accessibility analysis is performed to realize the division of the basic power supply units of the distribution network, and further generate the distribution network power supply map, providing structured basic data support for subsequent user group distribution analysis and electricity load forecasting.

[0035] S2: Based on the power distribution network diagram and historical mobile signaling data, perform behavioral pattern clustering to obtain the user group distribution of each basic power supply unit;

[0036] Anonymized historical mobile signaling data is obtained, including historical mobile signaling data from weekdays, holidays, and days of school closures and working from home. Based on date type and time period, hierarchical clustering based on dynamic time warping is performed on the historical mobile signaling data to generate user group clusters. These user group clusters include stable residential, regular office, public healthcare, education / learning, and business / entertainment groups. The total number of user group baselines is obtained by statistically analyzing the user group clusters according to date type and time period, and a user group conversion probability matrix is ​​constructed. The total number of user groups is calculated based on the total number of user group baselines and the user group conversion probability matrix. The distribution of the number of user groups within each basic power supply unit is fitted using a conditional Gaussian mixture model based on the power distribution network diagram.

[0037] Using weather, time period, and date type as conditional variables, a continuous-time Markov transition intensity matrix is ​​constructed, and the conversion probability between different user groups is statistically quantified. The parameters in the transition intensity matrix are trained by maximum likelihood estimation to obtain the user group conversion probability matrix.

[0038] Based on the historical mobile signaling data and user group clusters, a conditional Gaussian mixture model is used to fit the distribution of user group numbers within each basic power supply unit. Conditional variables are constructed using weather, time period, and date types. The conditional Gaussian mixture model is trained according to the user group clusters to obtain the distribution of user group numbers within each basic power supply unit under different conditional variables. The number of user groups within each basic power supply unit is calculated based on the conditional variables and the distribution of user group numbers.

[0039] The number of users within each basic power supply unit is calculated based on the condition variables and the user group size distribution, including: the user group size distribution formula is:

[0040] ;

[0041] In the formula, Basic power supply unit Inner Number of users in this category For the first Total number of users in this category Basic power supply unit , For condition variables Next User groups in basic power supply units The probability density function of occurrence.

[0042] It should be further explained that, in this embodiment, behavioral pattern clustering is performed based on the power distribution network diagram and historical mobile signaling data to obtain the user group distribution of each basic power supply unit, specifically including the following process.

[0043] First, anonymized historical mobile signaling data is obtained. This data originates from signaling collection on the communication network side and undergoes desensitization and anonymization processes during collection to ensure it does not contain any information that could identify a specific individual. The historical mobile signaling data is categorized by date type, including historical mobile signaling data for weekdays, holidays, and days with school closures or work-from-home arrangements. Specifically, historical mobile signaling data for weekdays reflects users' daily activity patterns under normal production and life rhythms; historical mobile signaling data for holidays reflects users' behavioral characteristics under conditions of increased leisure, travel, and family activities; and historical mobile signaling data for days with school closures or work-from-home arrangements reflects the characteristics of users' shrinking activity range and increased residential behavior under special social operating conditions.

[0044] After acquiring the historical mobile signaling data, it is preprocessed based on date type and time period. Specifically, the historical mobile signaling data is sliced ​​according to a preset time period division rule, converting the spatial residence and movement trajectory of each user at different time periods within a day into a time series format. The time period division is based on urban operational characteristics, divided into early morning, morning, daytime, evening peak, and nighttime periods, thus making the user behavior rhythms under different date types comparable. After completing the time series construction, hierarchical clustering analysis based on dynamic time warping is performed on the historical mobile signaling data. The time series of different users are aligned on the time axis using the dynamic time warping method to eliminate the differences in the timing of activities of different users. Then, the hierarchical clustering method is used to aggregate users based on the similarity of their activity rhythms under different date types and different time periods, generating several user clusters. Through the above clustering process, user clusters including stable residential, regular office, public medical, educational, and commercial / entertainment types are formed. Stable residential users tend to stay in their residential areas for extended periods with minimal fluctuations across time periods. Regular office users tend to concentrate in office areas during weekdays and exhibit clear commuting patterns in the morning and evening. Public healthcare users are distributed around medical facilities around the clock. Education and learning users concentrate in educational areas during teaching activities and their behavior patterns change significantly during holidays and school closures. Commercial and entertainment users concentrate in commercial and entertainment areas in the evenings and on holidays and exhibit strong time sensitivity.

[0045] After obtaining user clusters, statistics are performed on these clusters based on date type and time period to obtain the total baseline number of user groups under different date types and time periods. This total baseline number characterizes the typical scale of various user groups across the entire region under specific date types and time periods, serving as the basis for subsequent modeling of dynamic changes in user group numbers. Based on this, a user group conversion probability matrix is ​​constructed. This matrix depicts the probabilistic relationships of conversion between different user groups under different date types and time periods. For example, during holidays, some educational learning users are no longer concentrated in educational areas, and their behavioral characteristics shift to those of stable residential users; this conversion relationship is quantified using the user group conversion probability matrix.

[0046] In the specific implementation process, a continuous-time Markov transition intensity matrix is ​​constructed using weather, time period, and date type as conditional variables. The continuous-time Markov transition intensity matrix is ​​the mathematical foundation for describing the temporal evolution of user behavior, describing the instantaneous intensity of state transitions between different user groups over a continuous time scale. Instantaneous intensity is a probability density concept, representing the likelihood of a user transitioning from their current group to another group within a time window. Weather conditions include quantitative indicators such as temperature, humidity, and weather phenomena; time period conditions divide the day into intervals with different social activity characteristics, such as morning rush hour, lunch break, evening leisure, and late-night low-activity periods; date type conditions clearly distinguish between ordinary workdays, school closures and work-from-home days, regular weekends, and national statutory holidays. Statistical analysis is performed on user group transition events recorded in historical mobile signaling data. A transition event refers to the factual record of a change in the user group category of the same anonymous user within two adjacent analysis time periods. The maximum likelihood estimation method is used to iteratively train the undetermined parameters in the transition intensity matrix. The training process aims to maximize the probability of observed historical conversion event sequences. Parameter values ​​are continuously adjusted through an optimization algorithm until the output achieves optimal matching with real data, thus obtaining a stable and convergent user group conversion probability matrix. This matrix intuitively reflects the probability distribution of each user group converting to other user groups within a certain time scale, such as at 8 PM on a Sunday evening with a temperature of 30 degrees Celsius, under any given set of conditional variables. The matrix is ​​a square matrix where rows represent the initial conversion group and columns represent the target conversion group. Each element in the matrix represents a conditional probability between zero and one.

[0047] After obtaining the baseline total number of users and the user conversion probability matrix, the dynamic extrapolation calculation of the total number of users is performed. The baseline total number of users is derived from direct statistics of historical user group size under different date types and time periods, providing a static benchmark. Based on the baseline total number of users and the user conversion probability matrix, the dynamic total number of users under any target date type and target time period is calculated.

[0048] After calculating the total number of user groups, a conditional Gaussian mixture model is used to fit the distribution of user groups within each basic power supply unit based on the power distribution network diagram. Specifically, after obtaining the user group cluster division and constructing the basic power supply units, historical mobile signaling data is spatially correlated with user group clusters. This process first determines the spatial relationship between each location coordinate point in the anonymized user mobile signaling data and the polygonal geographical boundaries of each basic power supply unit in the power distribution network diagram generated in step S1. Geographic information algorithms, such as point-polygon inclusion judgment, are used to determine which basic power supply unit each signaling location point falls within. Simultaneously, the user corresponding to this location point is categorized into a specific user group cluster, such as stable residential or commercial / recreational. This mapping relationship links abstract user behavior categories with specific power grid geographic spatial structures. Probabilistic modeling of the spatial distribution characteristics of user groups is then performed. A multi-dimensional conditional variable is constructed using meteorological data, time period attributes, and date type. This conditional variable is a comprehensive input vector used to characterize the external environment and social clock state that influence people's travel and stay decisions at a specific moment. For example, an instance of a condition variable may contain specific information about three dimensions: hot weather, evening hours, and holidays.

[0049] Independent conditional Gaussian mixture models, each governed by the aforementioned conditional variables, are trained for different user groups. During training, for the stable residential user group, historical location data of all users classified as belonging to this group are extracted, and these location data are paired with the corresponding conditional variables at the time they were recorded, serving as training samples. The training algorithm learns a set of parameters that enable the model to output a probability density function in two-dimensional geographic space given any combination of conditional variables. This function intuitively describes the likelihood of a stable residential user appearing at any basic power supply unit in the city under specific conditions, such as a rainy morning rush hour on a weekday.

[0050] By completing the above training process for each user group cluster, a complete set of models is ultimately obtained. The models can output the spatial probability density function of each user group within each basic power supply unit under different conditional variables. The probability density function accurately quantifies the dynamic changes in population distribution with weather, time, and social calendar, serving as a crucial bridge between macro-level group behavior prediction and micro-level spatial unit load prediction. For example, "What is the expected density of commercial and entertainment-oriented people in a certain power supply unit in the city center on the night of the upcoming National Day holiday?" provides unprecedented data insights for the predictive and safe management of the power grid.

[0051] Based on this, the number of users within each basic power supply unit is calculated according to the condition variables and the distribution of user numbers. The calculation of the user number distribution is implemented as follows: For any basic power supply unit and any user group, the spatial range corresponding to the basic power supply unit is integrated according to the spatial probability density function of the user group under the constraints of the condition variables, and multiplied by the total number of users in the user group to obtain the number of users in the corresponding user group within the basic power supply unit.

[0052] Through the above steps, we have achieved refined clustering analysis of user group behavior patterns, dynamic modeling of user group size and its mutual transformation relationships, and quantitative calculation of the spatial distribution of user groups in each basic power supply unit, taking into full account the influence of date type, time period and meteorological conditions. This provides high-precision basic data support for subsequent distribution network load forecasting and operation safety analysis based on user group characteristics.

[0053] S3: Based on historical electricity consumption data, analyze the data according to weather, time period, and date type to obtain basic electricity consumption parameters for user groups;

[0054] Historical electricity consumption data of different user groups and electricity consumption data of infrastructure in different basic power supply units are obtained. Based on weather, time period and date type, condition variables are constructed to perform piecewise linear regression analysis on the historical electricity consumption data. Electricity load curves of different user groups and infrastructure are constructed. Based on the electricity load curves, the basic electricity load of individual users of different user groups and the basic electricity load of infrastructure in different power supply units are obtained.

[0055] It should be further explained that two types of long-term historical electricity consumption data are extracted from the distribution network management database. The first type of data is associated with different user groups identified through behavioral clustering in step S2. For example, for the business office user group, the historical electricity consumption data is an aggregated data sequence formed by aggregating and de-identifying the total electricity load data of users classified as this group in multiple typical commercial office areas during the same time period. The second type of data focuses on the electricity load data of infrastructure in different basic power supply units, such as street lighting, sewage treatment pumping stations, large communication switching hubs, and centralized charging stations for electric vehicles. The energy consumption of these infrastructures is usually independent of the load of ordinary users and has unique operating patterns, so they are analyzed as separate entities.

[0056] After data preparation, a structured conditional variable is constructed to characterize the impact of external environment and time cycles on electricity consumption behavior. This conditional variable consists of three dimensions: meteorological, time-of-day, and date-type. The meteorological dimension typically uses quantitative indicators such as temperature and relative humidity. The time-of-day dimension divides the 24 hours of a day into periods with social activity characteristics, such as evening hours when residents are more active and daytime hours when industrial production is at its peak. The date-type dimension clearly distinguishes between ordinary weekdays, Saturdays and Sundays, and national holidays, which represent different social rhythms. Any specific combination of these three dimensions, such as "a Sunday evening with a temperature exceeding 30 degrees Celsius," defines a unique and specific electricity consumption scenario.

[0057] Piecewise linear regression analysis is employed to deeply fit historical electricity consumption data for each user group and each infrastructure item. The advantage of this method lies in its ability to capture the non-globally linear, inflection-point-dependent relationships between electricity load and complex conditional variables. The analysis process first determines one or more threshold cutoff points for key influencing factors (such as temperature) based on statistical test results or domain prior knowledge. For example, the analysis reveals that residential electricity load increases linearly with decreasing temperature when outdoor temperature is below a certain threshold (heating effect), and increases linearly with increasing temperature when it is above another threshold (cooling effect), while maintaining a relatively stable baseline load within the comfort zone between the two thresholds. Based on these cutoff points, the historical dataset is automatically divided into several homogeneous sub-intervals. Within each sub-interval, using multidimensional conditional variables as input features and the corresponding historical average electricity load as the output target, a rigorous multiple linear regression calculation is performed to obtain a set of optimal regression coefficients within that local interval. By fully executing the above piecewise modeling process for each analytical object, a corresponding electricity load curve is constructed. Mathematically, this curve is represented as a piecewise linear function that can select an appropriate linear model based on any combination of weather, time period, and date types input to calculate the predicted typical load value.

[0058] From these electricity load curves, the essential electricity consumption parameters are extracted. For different user groups, the electricity load curve describes the total load of that group under specific conditions. To obtain parameters that reflect the typical behavior of a single individual and are independent of group size, the total load forecast needs to be divided by the standard statistical number of people in the group under those conditions, thereby resolving the basic electricity load of individual users for different user groups. For the basic electricity load of infrastructure in different power supply units, its load output value under standard reference condition variables is read, thus obtaining the basic electricity load of that facility.

[0059] S4: Obtain meteorological data for the future time period, combine user group distribution and basic power consumption parameters to predict the power load of each basic power supply unit and generate a distribution network load forecast map;

[0060] Future conditional variables are constructed based on date type and meteorological data within the future time period. The power load of each basic power supply unit is calculated based on the future conditional variables. A power distribution network load prediction map is generated based on the power distribution network diagram and the power load of each basic power supply unit.

[0061] Based on the number of user groups within each basic power supply unit, the basic electricity load per user of different user groups, and the basic electricity load of infrastructure in different power supply units, the electricity load of each basic power supply unit is calculated using the load forecasting calculation formula. The load forecasting calculation formula is as follows:

[0062] ;

[0063] In the formula, For the electrical load of the basic power supply unit, For condition variables Lower user groups The basic electricity load of a single user The electrical load of the infrastructure for the basic power supply unit.

[0064] It should be further explained that, through a data interface, high-precision, gridded weather forecast data for the target future time period is obtained in real time from authorized meteorological service centers. This data includes key information such as hourly temperature, humidity, wind speed, and weather phenomenon codes. Simultaneously, based on the target date's position in the calendar, its date type is determined, for example, whether it is a regular Tuesday, Saturday, or a statutory holiday like National Day. The hourly meteorological data for the future time period is combined with the corresponding date type to construct future condition variables used for forecasting. The dimensional definitions of these future condition variables are consistent with the condition variables used to analyze historical data in step S3, ensuring mathematical consistency and physical rationality from historical behavior patterns to future load projections.

[0065] After defining the future condition variables, the predicted electricity load for each basic power supply unit defined in the power supply diagram of the distribution network is calculated for each future time period. Specifically, the load calculation for any basic power supply unit i is implemented as follows: Based on the conditional Gaussian mixture model trained in step S2 and the constructed future condition variables, the number of various user groups expected to appear within the basic power supply unit under this specific future scenario is calculated. Using the same future condition variables, the electricity load curve constructed for each user group in step S3 is retrieved. From this curve, the basic electricity load per user of the user group under the condition variables is obtained; this value represents the average electricity load of a typical individual under this environment. Next, the total load contributed by all users within the basic power supply unit is calculated.

[0066] In addition to dynamic user loads, there are also relatively fixed infrastructure power consumptions within basic power supply units. Therefore, the calculation needs to include a constant term, namely the inherent basic power load of all infrastructure within the unit obtained from step S3. The inherent basic power load of the infrastructure is independent of the dynamic distribution of the user group. After calculating the loads of all basic power supply units for all future periods of interest according to the formula, spatial fusion of the data is performed. The distribution network power supply map generated in step S1, which includes the physical connection and regional division of the power grid, is used as the spatial base map. The calculated load values ​​are used as attribute data and accurately associated with each corresponding basic power supply unit polygon in the map. Using geographic information technology or a professional graphics engine, these load data are rendered on the base map in the form of color gradients, contour lines, or 3D bar charts, and finally synthesized to generate a dynamic distribution network load forecast map.

[0067] S5: Perform operational safety analysis based on the power distribution network load forecast diagram to generate early warning information, and assist operation and maintenance personnel in making countermeasure decisions based on the early warning information.

[0068] Based on the distribution network load forecast map, operational safety analysis is performed using a graph neural network. The graph structure is defined with basic power supply units as nodes and their connections as edges. Node feature vectors are constructed based on meteorological data for future periods, the power load of each basic power supply source, and the number of users within each basic power supply unit. A graph attention network is then used for operational safety analysis, generating early warning information that includes basic power supply units and topology. This early warning information assists maintenance personnel in making decisions regarding specific basic power supply units and topology.

[0069] It needs to be further explained that, such as Figure 2As shown, to achieve intelligent safety analysis, the distribution network load forecast diagram is transformed into a mathematical graph structure that can be processed by the computational model. Each basic power supply unit is treated as an independent node in the graph. The electrical connections between these nodes, such as the power supply paths formed by distribution lines or switching equipment, are abstractly defined as edges connecting these nodes. Thus, the entire distribution network forms a topology graph with basic power supply units as nodes and electrical connections as edges. This graph structure fully preserves the physical connection logic and energy transmission paths of the power grid.

[0070] After constructing the graph structure, a feature vector is built for each node, i.e., each basic power supply unit, to comprehensively characterize its future state. This node feature vector is a comprehensive description that integrates multi-dimensional information. It integrates refined meteorological data for the future time period, including the total power load of the basic power supply unit predicted in step S4, which is a direct indicator for assessing overload risk. The vector also incorporates the predicted distribution of the number of users within the unit, because different load compositions will affect its fluctuation characteristics and controllability. For example, a unit mainly occupied by commercial and entertainment users will have different load fluctuation characteristics than a unit mainly occupied by stable residential users. This information together constitutes the node's feature vector, serving as the raw input for deep analysis of the graph neural network.

[0071] A graph attention network model is employed for safety operation analysis. The core mechanism of this network lies in its attention layer, which adaptively learns the influence weights between any two connected nodes in the graph. During training, the network learns how, for example, when an upstream power node experiences extremely high load, the power supply capacity to multiple downstream load nodes will decrease proportionally; or how, when a region experiences a surge in load due to concentrated air conditioning startup, this pressure will be transmitted through the grid topology to its neighboring regions. By analyzing the weighted aggregation information of node characteristics and the characteristics of its neighboring nodes, the model can identify hidden risks that are difficult to detect using traditional thresholding methods, such as the cascading overload risk caused by multiple medium-load nodes supplying power through the same critical cross-section.

[0072] After deep computation and reasoning using graph attention networks, a comprehensive assessment of the future operating status of the distribution network is output. This assessment result is then transformed into specific early warning information. The early warning information clearly indicates the specific basic power supply unit numbers where high risks exist, such as "basic power supply units A-07 and B-12"; reveals the topological path where the risk is located, such as "10kV tie line L-5 connecting substations Alpha and Beta"; and indicates the nature and level of the risk, such as "the predicted load will reach 125% of the line's thermal stability limit, mainly due to the concentrated influx of commercial and entertainment loads during the evening peak."

[0073] Ultimately, structured and spatialized information is pushed to the management terminals of distribution network operators. Through early warning information, maintenance personnel shift their decision-making from experience-driven to data-driven. They can proactively deploy countermeasures for specific basic power supply units and topologies identified in the early warning information. For example, for critical weak topology paths, network operation modes can be adjusted, transferring some load to lightly loaded lines. This enhances the resilience and proactive management capabilities of the distribution network in the face of future complex operating conditions.

[0074] Historical electricity consumption data of different user groups and electricity consumption data of infrastructure in different basic power supply units are obtained. Based on weather, time period and date type, condition variables are constructed to perform piecewise linear regression analysis on the historical electricity consumption data. Electricity load curves of different user groups and infrastructure are constructed. Based on the electricity load curves, the basic electricity load of individual users of different user groups and the basic electricity load of infrastructure in different power supply units are obtained.

[0075] Future conditional variables are constructed based on date type and meteorological data for future time periods. The power load of each basic power supply unit is calculated based on the future conditional variables. A power distribution network load prediction map is generated based on the power distribution network diagram and the power load of each basic power supply unit.

[0076] The calculation of the electricity load of each basic power supply unit based on the aforementioned future condition variables includes: calculating the electricity load of each basic power supply unit using a load forecasting calculation formula based on the number of user groups within each basic power supply unit, the basic electricity load per user of different user groups, and the basic electricity load of infrastructure in different power supply units. The load forecasting calculation formula is as follows:

[0077] ;

[0078] In the formula, For the electrical load of the basic power supply unit, For condition variables Lower user groups The basic electricity load of a single user The electrical load of the infrastructure for the basic power supply unit.

[0079] Based on the distribution network load forecast map, operational safety analysis is performed using a graph neural network. The graph structure is defined with basic power supply units as nodes and their connections as edges. Node feature vectors are constructed based on meteorological data for future periods, the power load of each basic power supply source, and the number of users within each basic power supply unit. A graph attention network is then used for operational safety analysis, generating early warning information that includes basic power supply units and topology. This early warning information assists maintenance personnel in making decisions regarding specific basic power supply units and topology.

[0080] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A power grid operation safety analysis method based on multi-source data fusion and big data, characterized in that, Includes the following steps: S1: Collect distribution network topology data and distribution network area data, and based on the distribution network topology data and distribution network area data, divide the basic power supply units according to the distribution network topology nodes to generate a distribution network power supply diagram; S2: Based on the power distribution network diagram and historical mobile signaling data, perform behavioral pattern clustering to obtain the user group distribution of each basic power supply unit. This includes: obtaining anonymized historical mobile signaling data, which includes historical mobile signaling data for weekdays, holidays, and days with school closures and work-from-home arrangements; performing hierarchical clustering based on dynamic time warping on the historical mobile signaling data based on date type and time period to generate user group clusters, which include stable residential, regular office, public medical, educational, and commercial entertainment types; statistically analyzing user group clusters based on date type and time period to obtain the total number of user group baselines and constructing a user group conversion probability matrix; calculating the total number of user groups based on the total number of user group baselines and the user group conversion probability matrix; and fitting the distribution of user group numbers within each basic power supply unit using a conditional Gaussian mixture model based on the power distribution network diagram. The step of obtaining the total number of user group baselines and constructing a user group conversion probability matrix by statistically analyzing user group clusters based on date type and time period includes: constructing a continuous-time Markov transition intensity matrix with weather, time period and date type as condition variables, statistically quantifying the conversion probability between different user groups, and training the parameters in the transition intensity matrix through maximum likelihood estimation to obtain the user group conversion probability matrix. Based on the power distribution network diagram, the distribution of user groups within each basic power supply unit is fitted using a conditional Gaussian mixture model. This includes: fitting the distribution of user groups within each basic power supply unit using a conditional Gaussian mixture model based on the historical mobile signaling data and user group clusters; constructing conditional variables based on weather, time period, and date types; training the conditional Gaussian mixture model according to user group clusters; obtaining the distribution of user groups within each basic power supply unit under different conditional variables; and calculating the number of user groups within each basic power supply unit based on the conditional variables and the distribution of user groups. The formula for the distribution of user groups is: ； In the formula, Basic power supply unit Inner Number of users in this category For the first Total number of users in this category Basic power supply unit The total amount, For condition variables Next User groups in basic power supply units The probability density function of occurrence; S3: Based on historical electricity consumption data, analyze the data according to weather, time period, and date type to obtain basic electricity consumption parameters for user groups; S4: Obtain meteorological data for future periods, combine user group distribution and basic electricity consumption parameters to predict the electricity load of each basic power supply unit and generate a distribution network load forecast map; S5: Perform operational safety analysis based on the power distribution network load forecast diagram to generate early warning information, and assist operation and maintenance personnel in making countermeasure decisions based on the early warning information.

2. The power grid operation safety analysis method based on multi-source data fusion and big data as described in claim 1, characterized in that, The process of collecting distribution network topology data and distribution network area data, and dividing the distribution network into basic power supply units based on distribution network topology nodes to generate a distribution network power supply map, includes: obtaining distribution network topology data and distribution network area data from a distribution network management database; the distribution network area data including area function type, infrastructure, and area boundaries; performing power supply accessibility analysis based on the distribution network topology nodes in the distribution network topology data and the distribution network area data to divide the distribution network into basic power supply units; and then, based on the division... Each basic power supply unit generates a power distribution network diagram.

3. The power grid operation safety analysis method based on multi-source data fusion and big data as described in claim 1, characterized in that, The process of obtaining basic electricity consumption parameters for user groups by analyzing historical electricity consumption data according to weather, time period, and date type includes: acquiring historical electricity consumption data for different user groups and electricity consumption data for infrastructure in different basic power supply units; constructing conditional variables based on weather, time period, and date type to perform piecewise linear regression analysis on the historical electricity consumption data; constructing electricity load curves for different user groups and infrastructure; and obtaining the basic electricity load of individual users for different user groups and the basic electricity load of infrastructure in different power supply units based on the electricity load curves.

4. The power grid operation safety analysis method based on multi-source data fusion and big data as described in claim 1, characterized in that, The process of acquiring meteorological data for future periods, combining user group distribution and basic electricity consumption parameters to predict the electricity load of each basic power supply unit and generate a distribution network load forecast map includes: constructing future conditional variables based on date type and meteorological data for future periods; calculating the electricity load of each basic power supply unit based on the future conditional variables; and generating a distribution network load forecast map based on the distribution network power supply map and the electricity load of each basic power supply unit.

5. The power grid operation safety analysis method based on multi-source data fusion and big data according to claim 4, characterized in that, The calculation of the electricity load of each basic power supply unit based on the aforementioned future condition variables includes: calculating the electricity load of each basic power supply unit using a load forecasting calculation formula based on the number of user groups within each basic power supply unit, the basic electricity load per user of different user groups, and the basic electricity load of infrastructure in different power supply units. The load forecasting calculation formula is as follows: ； In the formula, Basic power supply unit The electrical load, For condition variables Next Basic electricity load per user for a user group of this type. Basic power supply unit The electricity load of infrastructure.

6. The power grid operation safety analysis method based on multi-source data fusion and big data according to claim 1, characterized in that, The system performs operational safety analysis based on the distribution network load forecast map to generate early warning information, and assists maintenance personnel in making response decisions based on the early warning information. This includes: performing operational safety analysis using a graph neural network based on the distribution network load forecast map; defining a graph structure based on the distribution network load forecast map, with basic power supply units as nodes and the connection relationships of basic power supply units as edges; constructing node feature vectors based on meteorological data in future time periods, the power load of each basic power supply unit, and the number of user groups within each basic power supply unit; performing operational safety analysis using a graph attention network; generating early warning information containing basic power supply units and topology structure; and assisting maintenance personnel in making response decisions for specific basic power supply units and topology structures based on the early warning information.

Images