Hierarchical classification dynamic cluster grouping method and system for air storage and charging resources in building park

Abstract

The invention discloses a hierarchical and classified dynamic cluster grouping method and system for air storage and charging resources in a building park, and relates to the technical field of resource allocation. According to the method, the double representative point design of the cluster center and the maximum dispersion point is adopted, the traditional inter-cluster distance calculation amount and complexity are greatly reduced, the calculation power demand and the operation cost are reduced, and the clustering efficiency is improved to adapt to a large-scale resource scene; the inter-cluster distance is measured by a median, cluster features are comprehensively represented by combining double representative points, deviated samples, data deviation and noisy points are effectively filtered, and even in a scene with multiple operation subject data and few samples, it can still be ensured that a clustering result is stable and accurate, and resource response characteristics are truly reflected; time sequence characteristics and dynamic iteration capability are considered, resource changes are adapted in real time, a scientific basis is provided for fine scheduling and optimal configuration, and dual improvement of resource utilization efficiency and management level is realized.

Description

Technical Field

[0001] This invention relates to the field of resource allocation technology, and in particular to a method and system for hierarchical classification and dynamic clustering of storage, filling and empty resources in building parks. Background Technology

[0002] With the advancement of "dual carbon" goals and the transformation of the energy structure, building industrial parks, as important units of energy consumption, urgently need to improve the intelligence and precision of their energy management. The large-scale integration of "storage, charging, and air conditioning" resources within these parks, such as energy storage systems, electric vehicle charging stations, and central air conditioning systems, while bringing enormous potential for flexible regulation, also presents unprecedented challenges to the operation and scheduling of building industrial parks. These resources are numerous, spatially dispersed, heterogeneous in characteristics, and their operating states change over time. How to efficiently aggregate and optimize their regulation is key to achieving cost reduction and efficiency improvement, safe and stable operation, and participation in grid interaction within the parks. Summary of the Invention

[0003] The purpose of this invention is to provide a hierarchical classification and dynamic clustering method and system for storage, charging and empty resources in building parks, in order to improve the technical problems of difficult resource management and low scheduling efficiency in building parks due to the large number of heterogeneous storage, charging and empty resources and their dynamic and time-varying characteristics. Traditional methods have problems such as clustering bias, computational complexity, fixed weights and insufficient real-time performance.

[0004] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:

[0005] A hierarchical classification and dynamic clustering method for storage, replenishment, and emptying resources in a building park, comprising:

[0006] Based on the geographical location and energy storage / charging / air conditioning resource types of the building park, the building park is classified into layers, and the actual operation data of the energy storage / charging / air conditioning equipment at each layer is collected to construct the corresponding three-dimensional characteristic vector; the energy storage / charging / air conditioning equipment is energy storage equipment, charging pile, or air conditioning.

[0007] By combining the three-dimensional characteristic vectors of each storage, charging and emptying device and the interactive scenario, the distance between each pair of different storage, charging and emptying devices at each level is calculated and aggregated through an adaptive distance weight allocation algorithm to generate an initial cluster;

[0008] Based on the cluster centers and maximum dispersion points of each initial cluster, the inter-cluster distances of each initial cluster are calculated using an improved agglomerative hierarchical clustering algorithm, and clustering iterations are performed to generate resource aggregation results.

[0009] In the above scheme, this method classifies building parks into hierarchical categories based on geographical location and resource type, accurately deconstructing the complex system of heterogeneous storage, charging, and emptying resources within the building parks. It focuses on the aggregation needs of similar resources, solving the management and scheduling chaos caused by the dispersion and large differences in characteristics of heterogeneous resources. Relying on three-dimensional characteristic vectors to characterize the core attributes (power, response time, cost) of storage, charging, and emptying equipment, and combining an adaptive distance weight allocation algorithm to adapt to the differentiated needs under different interaction scenarios, it avoids the drawback of fixed weights being unable to match dynamic scheduling objectives. By leveraging the improved agglomerative hierarchical clustering features of cluster centers and maximum dispersion points, it significantly reduces the computational complexity of traditional clustering and enhances the anti-interference ability against data bias and noise, solving the problems of low computational efficiency and easy distortion of clustering results in traditional algorithms. Through cluster iteration, it achieves dynamic tracking of resource status, effectively solving the problem of the disconnect between static clustering and the time-varying characteristics of resources. Finally, with accurate and efficient resource aggregation results, it provides reliable support for the refined scheduling and optimized allocation of storage, charging, and emptying resources in building parks, significantly improving resource utilization efficiency and grid interaction response capabilities.

[0010] Further, the generation of the initial cluster includes:

[0011] Each storage, charging, and emptying device at each level is used as a cluster point, and basic weights are set for different types of storage, charging, and emptying devices; the basic weights include power basic weight, response time basic weight, and cost basic weight;

[0012] Set adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment;

[0013] Based on the three-dimensional characteristic vectors and interactive scenarios of each storage, charging and emptying device, the scenario correction coefficient for each level is calculated through the scenario judgment function;

[0014] Based on the scene correction coefficient and state correction coefficient of each level, update the basic weights of each level and generate the corresponding final weights.

[0015] Based on the final weights of each level, the distance between any two different clusters at the same level is calculated and aggregated to obtain the corresponding initial clusters.

[0016] In the above scheme, this method accurately configures basic weights in three dimensions for different types of energy storage, charging, and air conditioning equipment, aligning with the essential differences in the attributes of various types of equipment. This solves the problem of traditional clustering weight settings being one-size-fits-all and unable to adapt to the characteristics of heterogeneous equipment. By combining the three-dimensional characteristic vector of the equipment and introducing adjustment coefficients and weight benchmark calculation state correction coefficients, dynamic adaptation to the real-time operating status of the equipment (such as energy storage SOC, charging pile user dwell time, and air conditioning temperature deviation) is achieved, avoiding distance calculation distortion caused by the disconnect between static weights and the actual state of the equipment. Relying on the scenario judgment function and combining the interactive scenario to calculate the scenario correction coefficient, the weights can accurately match differentiated scheduling needs such as frequency regulation, peak shaving, and economic scheduling, thus solving the problem of fixed weights. The problem of being unable to respond to changes in the scene is addressed. By dynamically updating the basic weights through scene correction coefficients and state correction coefficients to generate the final weights, the comprehensiveness and real-time nature of the weights are ensured, and the accuracy of distance calculation for storage, charging, and emptying equipment at the same level is improved. Based on the final weights, clustering points are aggregated to obtain initial clusters, which not only ensures the rationality and relevance of the initial clusters, but also provides a high-quality data foundation for subsequent improvements in hierarchical clustering. This effectively solves the problems of cluster division deviation and low efficiency of subsequent clustering caused by the coarse weight settings and neglect of equipment status and scene differences in traditional initial clustering. It further consolidates the accuracy and reliability of the overall resource aggregation results and provides solid initial clustering support for the refined scheduling of storage, charging, and emptying resources in building parks.

[0017] Further, the calculation of the state correction coefficient for each level includes:

[0018] When the When the energy storage and charging equipment at each level is an energy storage device, the calculation is based on the adjustment coefficient and weighting benchmark of the energy storage device and the corresponding state of charge of each energy storage device. State correction coefficients for each level of energy storage equipment;

[0019] When the When the energy storage and charging equipment at each level is a charging pile, calculate the charging-dwell time ratio and behavioral response level of each charging pile; based on the charging-dwell time ratio, behavioral response level, and corresponding adjustment coefficients and weighting benchmarks of each charging pile, calculate the first... Status correction coefficients for each charging pile at each level;

[0020] When the When the storage and charging equipment at each level is an air conditioner, the indoor temperature and user-set temperature of each air conditioner are collected, and the temperature deviation value of each air conditioner is calculated based on the set maximum temperature deviation; based on user comfort requirements, the temperature deviation value of each air conditioner, the adjustment coefficient, and the weighting benchmark, the first... The status correction coefficients for each air conditioner at each level.

[0021] In the above scheme, this method customizes exclusive calculation logic for the core characteristic differences of three types of equipment: energy storage, charging piles, and air conditioners. It accurately focuses on the key operating state parameters of each type of equipment: energy storage equipment focuses on the state of charge (SOC), charging piles focus on the charging-dwell time ratio and user behavior response, and air conditioners focus on temperature deviation and user comfort requirements. This solves the problem of traditional correction coefficient calculation being one-size-fits-all and ignoring the core state influencing factors of heterogeneous equipment. By introducing equipment-specific adjustment coefficients and weight benchmarks, and combining real-time collected equipment operating data (such as energy storage SOC and air conditioner indoor temperature) for dynamic calculation, it achieves accurate matching between correction coefficients and actual equipment operating states, avoiding weight distortion caused by the disconnect between static correction logic and the time-varying characteristics of equipment. The calculation dimensions are designed differently for the operating mechanisms of different equipment, so that state correction not only fits the essential attributes of the equipment, but also captures state fluctuations in real time. This solves the problems of distance calculation deviation and unreasonable cluster division caused by traditional methods that do not specifically consider the influencing factors of equipment state.

[0022] Furthermore, the calculation of the scene correction coefficient for each level using the scene judgment function includes:

[0023] Set frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold to obtain the current electricity price and grid frequency deviation at each level;

[0024] Collect historical operational data of the building park and use machine learning or deep learning algorithms to make predictions to obtain the load forecast value at the current moment;

[0025] Based on the frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold, the load forecast, electricity price, and grid frequency deviation at each level are judged by the scenario judgment function to determine the interaction scenario at each level.

[0026] Based on the interactive scenarios at each level, corresponding scenario correction coefficients are generated.

[0027] Furthermore, the formula corresponding to the scene judgment function is:

[0028] ;

[0029] in, Represents the maximum value function. Describes the minimum value function. This indicates the current system load status of the building complex. This represents the scene judgment function. , These represent the power grid frequency deviation and the frequency regulation action threshold, respectively. Represents absolute value. , Representing time respectively The load forecast values ​​for the building park and the set peak-valley difference threshold, , Representing time respectively Electricity price and electricity price fluctuation threshold, , , These represent frequency modulation scenarios, peak shaving scenarios, and economic dispatch scenarios, respectively.

[0030] In the above scheme, this method constructs a multi-dimensional scenario judgment basis by setting three types of core thresholds and combining real-time grid frequency deviation, electricity price data, and load forecast values. This solves the problem of scheduling demand matching deviation caused by traditional scenario adaptation relying on only a single indicator and biased judgment. Based on the scenario judgment function containing maximum and minimum value functions, a standardized scenario recognition logic is established, avoiding the drawbacks of scenario confusion and switching lag under manual judgment or fuzzy rules, and realizing accurate and automatic judgment of three types of core interactive scenarios. Based on the judgment results, a dedicated scenario correction coefficient is generated, enabling the weight allocation to respond to the dynamic scheduling needs of the grid in real time, solving the problem that traditional fixed scenario adaptation cannot keep up with the real-time changes in grid frequency, load, and electricity price. By predicting the energy consumption trend of the park in advance through load forecasting, the scenario correction is forward-looking, avoiding the scenario adaptation lag problem caused by relying solely on real-time data, further improving the synergy between resource clustering and scheduling decision-making, and effectively supporting the efficient participation of building park storage, charging, and empty resources in grid interaction.

[0031] Furthermore, the generated resource aggregation result includes:

[0032] Select any level and use the corresponding initial cluster as the starting point for iteration;

[0033] Determine the cluster center of each initial cluster and use it as the first representative point. Calculate the distance between different points in each initial cluster and the cluster center. Select the point with the largest dispersion in each initial cluster and use it as the second representative point.

[0034] Calculate the distances between the first and second representative points in each pair of different initial clusters and sort them from largest to smallest. Select the distance that is the median as the inter-cluster distance between the two corresponding initial clusters.

[0035] Iterate through all inter-cluster distances, select the pair of clusters with the smallest distance and merge them to generate a new cluster and iterate until the preset conditions are met to obtain the resource aggregation result of the selected level;

[0036] Reselect any level and iterate until you obtain the resource aggregation results for all levels.

[0037] In the above scheme, this method adopts a dual representative point design of cluster center and maximum dispersion point, which covers both the core characteristics of the cluster and the boundary features, solving the problem that the traditional single representative point is one-sided in representing cluster characteristics and causes distortion of inter-cluster distance measurement. By calculating the distance between the four sets of representative points of each pair of initial clusters and taking the median as the inter-cluster distance, the median's resistance to extreme values ​​and noise points effectively filters out data bias interference, solving the problem that the traditional full-point distance calculation is easily affected by deviated samples and has poor result stability. At the same time, the amount of inter-cluster calculation is reduced from the traditional n×m times to a fixed 4 times, which greatly reduces the computational complexity. This approach improves clustering efficiency, ensures the rationality of cluster merging and the orderliness of the clustering process, and avoids the problems of chaotic iterative logic and irregular cluster aggregation in traditional clustering. Through the whole-process design, it not only ensures that the resource aggregation results can truly reflect the behavioral similarity of storage, charging and emptying equipment at each level, but also takes into account computational efficiency and anti-interference ability. It effectively solves the core pain points of traditional clustering algorithms, such as large computational load, easy distortion of results, and poor adaptability. It provides accurate and efficient aggregation support for the refined scheduling of resources at each level of building parks and the interactive response of the power grid, further improving resource utilization efficiency and power grid operation stability.

[0038] A hierarchical and dynamic clustering system for storage, replenishment, and emptying resources in a building park, comprising:

[0039] The feature vector module is used to classify building parks into different levels based on their geographical location and the type of energy storage, charging, and air conditioning resources, and to collect the actual operating data of the energy storage, charging, and air conditioning equipment at each level to construct the corresponding three-dimensional feature vector; the energy storage, charging, and air conditioning equipment can be energy storage equipment, charging piles, or air conditioners.

[0040] The distance calculation module is used to combine the three-dimensional characteristic vectors of each storage, charging and emptying device with the interactive scene, and calculate the distance between each pair of different storage, charging and emptying devices at each level through an adaptive distance weight allocation algorithm and aggregate them to generate an initial cluster.

[0041] The clustering module is used to calculate the inter-cluster distance of each initial cluster based on the cluster center and the maximum dispersion point of each initial cluster, and to perform clustering iteration to generate resource aggregation results.

[0042] Furthermore, the distance calculation module includes:

[0043] The weight preset unit is used to set the basic weights of different types of storage, charging and emptying devices as cluster points at each level; the basic weights include power basic weight, response time basic weight and cost basic weight.

[0044] The first coefficient calculation unit is used to set the adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and to calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment.

[0045] The second coefficient calculation unit is used to calculate the scene correction coefficient of each level based on the three-dimensional characteristic vectors and interactive scenes of each storage, filling and emptying device through the scene judgment function.

[0046] The weight update unit is used to update the basic weights of each level based on the scene correction coefficient and state correction coefficient of each level, and generate the corresponding final weights.

[0047] The distance calculation unit is used to calculate and aggregate the distances between pairs of different clusters at the same level based on the final weights of each level, thus obtaining the corresponding initial clusters.

[0048] In the above scheme, this system uses a feature vector module to classify resources hierarchically according to geographical location and resource type. Combined with three-dimensional feature vectors, it comprehensively describes the core attributes of equipment, solving the problems of scattered heterogeneous resources and incomplete feature description. The distance calculation module has clear division of labor among its units, and the basic weights are adapted to the essential attributes of different equipment to avoid a one-size-fits-all approach to weights. The state correction coefficient is aligned with the real-time operating status of the equipment, and the scenario correction coefficient matches the differentiated needs of the power grid. The weight update generates accurate final weights, thereby accurately calculating the distance between equipment and aggregating the initial clusters. This solves the pain points of traditional static weights, poor scenario adaptation, and distorted distance calculations. Combined with the improved algorithm of the clustering module, the system achieves efficient and accurate resource aggregation, providing reliable support for the refined scheduling of storage, charging, and emptying resources in the park. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention;

[0051] Figure 2 This is a system structure diagram of Embodiment 1 of the present invention;

[0052] Figure 3 This is a resource aggregation result diagram of the energy storage device in Embodiment 2 of the present invention;

[0053] Figure 4 This is a time-period variation graph of the resource aggregation results of the energy storage device in Embodiment 2 of the present invention;

[0054] Figure 5 This is a diagram showing the resource aggregation results of charging piles in Embodiment 2 of the present invention;

[0055] Figure 6This is a time-period variation graph of the resource aggregation results of charging piles in Embodiment 2 of the present invention;

[0056] Figure 7 This is a diagram showing the resource aggregation results of the air conditioner in Embodiment 2 of the present invention;

[0057] Figure 8 This is a graph showing the cross-period changes in the resource aggregation results of the air conditioner in Embodiment 2 of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0059] Example 1:

[0060] Please see Figure 1 This embodiment provides a hierarchical classification and dynamic clustering method for storage, charging, and emptying resources in building parks. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0061] Due to the large number of heterogeneous resources such as energy storage devices, charging piles, and air conditioners in building parks, their dynamic characteristics can easily lead to problems of "difficult management and low scheduling efficiency." Therefore, this embodiment adopts the approach of "first hierarchical classification, then clustering, and dynamic updating." That is, the building park is first hierarchically classified according to its geographical location, then the park resources are classified according to resource type, and then in each cluster, dynamic clustering is performed based on the real-time status of the resources. This aggregates the park's adjustable resources into a few distinctive park adjustable resource clusters, thereby greatly reducing scheduling complexity and achieving accurate resource matching and efficient resource utilization.

[0062] Therefore, a hierarchical classification and dynamic clustering method for storage, replenishment, and empty space resources in building parks includes:

[0063] S1. Based on the geographical location and energy storage / charging / air conditioning resource types of the building park, the building park is classified into layers, and the actual operation data of the energy storage / charging / air conditioning equipment at each layer is collected to construct the corresponding three-dimensional characteristic vector; the energy storage / charging / air conditioning equipment is energy storage equipment, charging equipment (charging piles), or air conditioning.

[0064] S1 includes:

[0065] S1-1. Based on the geographical location of the building park, the building park is divided into different building sub-parks;

[0066] S1-2. Based on the type of energy storage, charging, and air conditioning resources in each building sub-park, classify each building sub-park and determine the energy storage, charging, and air conditioning equipment at each level; that is, in each building sub-park, classify each building sub-park according to the type and characteristics of energy storage equipment, charging piles, and air conditioning, and each level of the classification includes only energy storage equipment or charging equipment (charging piles) or air conditioning.

[0067] Specifically, within building parks, the resources covered by the corresponding level of energy storage devices (ESS) possess bidirectional power regulation capabilities. Through coordinated control of charging and discharging strategies, energy time-shifting can be achieved to better match energy supply and demand, providing a more flexible operating mode. Based on changes in grid demand, it dynamically adjusts its own power output or consumption, providing ancillary service support capabilities during peak and off-peak periods to maintain stable system operation. Electric vehicle charging piles (EVCPs) are classified as a separate layer because their fundamental physical characteristics, operating modes, and user behaviors differ significantly from energy storage devices and air conditioners. Charging times range from 15 minutes (fast charging) to 10 hours (slow charging), entirely depending on user needs.

[0068] The corresponding levels of air conditioning (AC) equipment include transferable loads with flexible adjustment capabilities during operation, transferable loads that support adjustment during operation, and interruptible loads that allow compression of operation time.

[0069] S1-3. Collect and standardize the actual operating data of the storage, filling and emptying equipment at each level, and construct a three-dimensional characteristic vector for each level.

[0070] Specifically, the actual operating data of different energy storage, charging, and air conditioning devices vary. For example, the actual operating data of energy storage devices includes real-time charging and discharging power, state of charge, charging and discharging efficiency, unit energy consumption cost, and response delay time; the actual operating data of charging piles includes real-time charging power, user's expected stay time, charging duration, remaining charging demand, unit charging cost, and user's allowable adjustment willingness; the actual operating data of air conditioners includes real-time operating power, actual indoor temperature, user-set temperature, maximum allowable temperature deviation, unit operating energy consumption cost, and load adjustment delay.

[0071] The three-dimensional characteristic vector includes the power characteristics, price, and response duration of different energy storage and charging devices. For example, the power characteristics of energy storage devices... Its price cost is due to its bidirectional power adjustment range and power response speed. The overall cost per unit charge / discharge, and its response duration. For sustainable adjustment duration based on SOC (State of Charge); power characteristics of charging piles The price cost is related to the peak power and power adjustment range of unidirectional charging. User-acceptable cost per unit of charge, and its response time The adjustable window duration (user dwell time - necessary charging time) is designed to avoid affecting user vehicle use; the power characteristics of the air conditioner. The price cost is related to the load adjustment range and load response speed. The energy cost per unit load regulation, and its response duration. This refers to the duration of sustainable adjustment based on temperature deviation (the longest duration during which it can be interrupted / transferred, provided that it does not exceed the maximum allowable temperature deviation).

[0072] The power characteristics of different energy storage, charging, and air conditioning devices need to be standardized because there are significant power differences between energy storage devices, air conditioners, and charging piles. For example, during the same time period, a high-load air conditioner corresponds to a power of 500kW, while a low-power charging pile corresponds to a power of only 7kW. When calculating the similarity distance, 7kW is negligible compared to the square of 500kW, causing the clustering results to simply group by power magnitude rather than by power consumption pattern. Therefore, the power characteristics need to be standardized before calculating the distance. Taking energy storage devices as an example, the specific steps for standardization are as follows:

[0073] If there are m energy storage devices at a certain level in a certain building complex, then the... The power of an energy storage device during a certain period of time is According to the formula:

[0074] ;

[0075] ;

[0076] Calculate the mean value of the power characteristics corresponding to this level. and standard deviation .in, This represents the summation function.

[0077] The mean value corresponding to the power characteristics of this level and standard deviation The power characteristics of this level are standardized, and the corresponding formula is:

[0078] ;

[0079] in, Indicates the first The power characteristics of a standardized energy storage device.

[0080] By integrating the price cost, response duration, and standardized power characteristics of each level, a three-dimensional characteristic vector is generated for each level.

[0081] S2. Combining the three-dimensional characteristic vectors of each storage, charging and emptying device with the interactive scene, the distance between each pair of different storage, charging and emptying devices at each level is calculated and aggregated using an adaptive distance weight allocation algorithm to generate an initial cluster;

[0082] For each level of storage, charging, and emptying equipment, processing is performed, treating each level of equipment as a cluster point. Based on different grid demands, weights are assigned to the different response characteristics of the resources, allowing the clustering results to better adapt to grid dispatch requirements. The weight factors are defined based on the characteristics of the storage, charging, and emptying resources in the park, and are adjusted in real time according to operating status and interaction scenario requirements. Therefore, S2 includes:

[0083] S2-1. Take each storage, charging, and emptying device at each level as a cluster point and set basic weights for different types of storage, charging, and emptying devices; the basic weights include power basic weight, response time basic weight, and cost basic weight.

[0084] Specifically, according to the formula:

[0085] ;

[0086] ;

[0087] ;

[0088] Set corresponding basic weights for each level of energy storage devices, charging devices (charging piles), or air conditioners. , , These represent the basic weights of energy storage devices, charging devices, and air conditioners, respectively. , , These represent the power-based weight, response time-based weight, and cost-based weight when a certain level of energy storage, charging, and emptying equipment is considered an energy storage device. , , These represent the power base weight, response time base weight, and cost base weight when a certain level of energy storage and charging equipment is used as a charging pile. , , These represent the power base weight, response time base weight, and cost base weight when the storage, charging, and air conditioning equipment at a certain level is an air conditioner.

[0089] S2-2. Set adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment.

[0090] Specifically, the state correction coefficients include power state correction coefficients, response time state correction coefficients, and cost state correction coefficients.

[0091] When the When the energy storage and charging equipment at each level is an energy storage device, the calculation is based on the adjustment coefficient and weighting benchmark of the energy storage device and the corresponding state of charge of each energy storage device. State correction coefficients for each level of energy storage devices The corresponding formula is:

[0092] ;

[0093] in, , , They represent the first The first level Power state correction factor, response time state correction factor, and cost state correction factor for each energy storage device. , , They represent the first The first level The adjustment coefficients corresponding to the power, response time, and cost of an energy storage device. , , They represent the first The first level The weighted benchmarks for the power, response time, and cost of each energy storage device. Indicates the first The first level The state of charge of an energy storage device.

[0094] When the When the energy storage and charging equipment at each level is a charging pile, calculate the charging-dwell time ratio and behavioral response level of each charging pile; based on the charging-dwell time ratio, behavioral response level, and corresponding adjustment coefficients and weighting benchmarks of each charging pile, calculate the first... Status correction coefficients for each charging station at each level The corresponding formula is:

[0095] ;

[0096] ;

[0097] in, , , They represent the first The first level The power state correction factor, response time state correction factor, and cost state correction factor for each charging pile. , They represent the first The first level The adjustment coefficient corresponding to the response time and cost of each charging pile. , They represent the first The first level The weighting benchmarks corresponding to the response time and cost of each charging station. , They represent the first The first level The charging-dwell time ratio and behavioral response level of each charging station , They represent the first The first level Users of each charging station are allowed to adjust the charging time window, and the total time they expect their vehicle to be connected to the charging station. Behavioral responsiveness level. It is used to quantify the responsiveness of user charging behavior to changes in electricity prices, and can be expressed as the ratio of the number of times a user responds to a price signal to the number of times the system issues a price change signal.

[0098] When the When the storage and charging equipment at each level is an air conditioner, the indoor temperature and user-set temperature of each air conditioner are collected, and the temperature deviation value of each air conditioner is calculated based on the set maximum temperature deviation; based on user comfort requirements, the temperature deviation value of each air conditioner, the adjustment coefficient, and the weighting benchmark, the first... Status correction coefficients for each level of air conditioning The corresponding formula is:

[0099] ;

[0100] ;

[0101] in, , , They represent the first The first level The power state correction factor, response time state correction factor, and cost state correction factor for each air conditioner. , They represent the first The first level The adjustment coefficients corresponding to the power and cost of each air conditioner. , They represent the first The first level The weighted benchmarks corresponding to the power and cost of each air conditioner. , They represent the first The first level The user comfort requirements and temperature deviation values ​​for each air conditioner. , , They represent the first The first level The indoor temperature, user-set temperature, and maximum temperature deviation of each air conditioner. It represents the absolute value. The value ranges from 0 to 1. The larger the value, the higher the user's comfort requirements and the more sensitive they are to temperature fluctuations. Therefore, the cost weight of scheduling will be very high.

[0102] S2-3. Based on the three-dimensional characteristic vectors and interactive scenarios of each storage, charging and emptying device, the scenario correction coefficients of each level are calculated through the scenario judgment function;

[0103] Because the power grid's resource allocation within the building complex varies at different times and under different operating conditions, this embodiment involves three interactive scenarios: frequency regulation scenario. Peak shaving scenarios and economic scheduling scenarios The scenario correction coefficients include power scenario correction coefficients, response time scenario correction coefficients, and cost scenario correction coefficients.

[0104] The formula corresponding to the scene judgment function is:

[0105] ;

[0106] in, Represents the maximum value function. Describes the minimum value function. This represents the current system load status of the building complex and serves as the input variable for the scenario judgment function. This represents the scene judgment function. , These represent the power grid frequency deviation and the frequency regulation action threshold, respectively. , Representing time respectively The load forecast values ​​for the building park and the set peak-valley difference threshold, , Representing time respectively Electricity price and electricity price fluctuation threshold.

[0107] Therefore, S2-3 includes:

[0108] S2-3-1. Set the frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold, and obtain the current electricity price and grid frequency deviation at each level.

[0109] S2-3-2. Collect historical operational data of the building park (historical load, time and weather data) and use machine learning or deep learning algorithms to make predictions to obtain the load forecast value at the current moment;

[0110] Specifically, statistical, machine learning, or deep learning models are used for training, and historical operational data (historical load, time, and weather data) collected from the building park are input into the trained machine learning or deep learning model for prediction. The final output is a predicted value of the future load, which will be directly used in the scenario judgment function to achieve advanced perception and decision-making of the system state.

[0111] In this embodiment, the LightGBM model is selected as the load forecasting model. First, the time characteristics (accurate to the hour, whether it is a weekday / holiday, and the peak / valley / normal period of the power grid) and weather characteristics (real-time temperature, deviation of the current day's temperature from the historical average) are converted into structured data recognizable by the model according to preset rules, and then input into the trained model. The model, relying on its built-in gradient boosting decision tree ensemble structure, automatically captures the intraday peak and trough of load, the intraweekly periodic variation patterns, and the sensitive correlation between temperature fluctuations and air conditioning load. It accurately maps the nonlinear correspondence between multi-dimensional input features and future load. Through the integrated voting and weighted calculation of multiple decision trees, it quickly outputs the predicted future load value corresponding to the current moment, providing accurate and advanced data support for the scenario judgment function, and helping the system achieve early perception and efficient decision-making regarding the power grid status.

[0112] S2-3-3: Based on the frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold, the load forecast value, electricity price, and grid frequency deviation of each level are judged by the scenario judgment function to determine the interaction scenario of each level.

[0113] S2-3-4. Based on the interactive scenarios at each level, generate corresponding scenario correction coefficients.

[0114] In this embodiment, if the storage, charging, and emptying equipment is in a frequency regulation scenario, the scenario correction coefficient is (1.5, 0.8, 0.7); if the storage, charging, and emptying equipment is in a peak shaving scenario, the scenario correction coefficient is (0.8, 1.3, 1.9); if the storage, charging, and emptying equipment is in an economic dispatch scenario, the scenario correction coefficient is (0.8, 0.9, 1.3).

[0115] S2-4. Based on the scene correction coefficient and state correction coefficient of each level, update the basic weights of each level and generate the corresponding final weights.

[0116] Specifically, taking the first Level 1 Taking a storage and charging device as an example, according to the formula:

[0117] ;

[0118] Calculate the updated weights ;in, Indicates the basic weight. Indicates the first Level 1 The state correction factor for each storage, charging, and emptying device. Indicates the first Level 1 The scenario correction coefficient for individual storage and charging equipment. This indicates element-wise multiplication.

[0119] Then normalize the updated weights to obtain the first... Level 1 The final weight of each storage and charging equipment.

[0120] S2-5. Based on the final weights of each level, calculate the distance between any two different clusters at the same level and aggregate them to obtain the corresponding initial clusters.

[0121] Specifically, taking the first Level 1 One storage and charging equipment and the first Taking a storage and charging equipment as an example, the formula corresponding to the distance is:

[0122] ;

[0123] in, Indicates the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment The square of the distance between them (distance > 0). , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment The final power weight, , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment Power characteristics, , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment The final weight of response time , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment Response duration, , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment Ultimate cost weighting , They represent the first Level 1 Individual storage and air filling equipment and the Individual storage and air filling equipment Price and cost.

[0124] The first Each cluster point corresponding to a storage, charging, and emptying device within a hierarchy serves as an independent initial unit. Subsequently, the pairwise distances of all initial units are traversed, and the pair of cluster points with the smallest distance (i.e., the two devices with the most similar response characteristics) are selected and merged to form a set containing these two cluster points. This process of selecting the smallest distance pair and merging is repeated until all unmerged independent cluster points within that hierarchy have completed their first aggregation (or merged into a preset number of initial clusters). The resulting set containing multiple similar cluster points is the initial cluster corresponding to that hierarchy. Each initial cluster represents a group of storage, charging, and emptying devices with similar response characteristics, providing the basic unit for subsequent inter-cluster iterative aggregation.

[0125] S3. Based on the cluster center and maximum dispersion point of each initial cluster, the inter-cluster distance of each initial cluster is calculated by an improved agglomerative hierarchical clustering algorithm, and clustering iteration is performed to generate resource aggregation results.

[0126] S3 includes:

[0127] S3-1. Select any level and use the corresponding initial cluster as the starting point for iteration;

[0128] S3-2. Determine the cluster center (mean point) of each initial cluster and use it as the first representative point. Using the same distance calculation method as S2, calculate the distance between all points in each initial cluster and the cluster center. Select the point with the largest dispersion in each initial cluster and use it as the second representative point.

[0129] S3-3. Calculate the distance between the first representative point and the second representative point in each pair of different initial clusters, sort them from largest to smallest, and select the distance as the median as the inter-cluster distance between the two corresponding initial clusters.

[0130] Specifically, with two initial clusters , For example, cluster Contains n points, cluster Given m points, calculate the distance and cluster between two first representative points using the same method as S2. First representative point and cluster Distance between the second representative points, cluster The second representative point and cluster The distance between the first representative points and the distance between the two second representative points are sorted in descending order. The median (or the average of the two medians if the total number is even) is selected as the inter-cluster distance between the two initial clusters.

[0131] S3-4. Traverse all inter-cluster distances, select the pair of clusters with the smallest distance and merge them to generate a new cluster;

[0132] S3-5. Repeat S3-2 to S3-4 until the preset conditions are met to obtain the resource aggregation results of the selected level. In this embodiment, the preset conditions are that the minimum distance between all clusters is greater than or equal to the preset threshold or the number of iterations reaches the preset upper limit.

[0133] S3-6. Repeat S3-1 to S3-5 until the resource aggregation results of all levels are obtained.

[0134] Traditional clustering algorithms suffer from high computational complexity, low efficiency, and high cost. They require calculating pairwise distances between all points in each cluster. Contains n points, cluster If there are m points, then nm distance calculations are required. The computational load increases exponentially with the sample size, demanding extremely high computing power and increasing hardware deployment and operating costs. Furthermore, it suffers from weak anti-interference capabilities and poor clustering accuracy. Many algorithms rely on mean statistics or direct calculation of the full distance, making them susceptible to deviations from the sample, data bias, and noise. This is particularly problematic in scenarios where park optical storage and charging resource data comes from multiple operators, has varying data quality, or has a small sample size, leading to biased clustering results that fail to reflect the true response characteristics of the resources. Some algorithms also exhibit one-sided representation, using only a single representative point to represent cluster characteristics while ignoring cluster boundary characteristics, resulting in distorted inter-cluster distance measurements. In contrast, the improved median clustering algorithm (improved median distance method) in this embodiment has significant advantages. It employs a dual-representative point design, using the cluster center (mean point) and the maximum dispersion point, requiring only the calculation of four sets of representative points between two clusters (the distance between the two first representative points, the cluster center, the mean point, and the maximum dispersion point). First representative point and cluster Distance between the second representative points, cluster The second representative point and cluster The distance between the first representative point and the distance between the two second representative points is used to reduce the computational complexity of the traditional algorithm from nm times to a fixed 4 times, significantly reducing computational complexity, improving clustering speed, reducing computing power requirements, and lowering operating and hardware costs. At the same time, the median is used as the final metric for inter-cluster distance. The median is not affected by extreme values, off-samples, and noise. Combined with the dual representative points, the central and boundary characteristics of the clusters are fully represented, effectively solving the problems of one-sided representation and distorted measurement in the traditional algorithm. It is especially suitable for park scenarios with multiple operating entities and small sample sizes. Through strong anti-interference capabilities and comprehensive representation capabilities, it ensures that the clustering results are stable and accurate, and can truly reflect the response characteristics of storage, charging, and empty resources. It provides reliable support for the dynamic clustering and efficient scheduling of storage, charging, and empty resources in the subsequent building park.

[0135] In this embodiment, the building park is clustered using a dynamic clustering method every 5 minutes.

[0136] In summary, this invention combines a hierarchical resource partitioning logic with an improved median distance-based agglomerative hierarchical clustering algorithm, providing precise and reliable technical support for the efficient management and scheduling of storage, charging, and emptying resources in building industrial parks. Its innovative use of a dual-representative point design ("cluster center + maximum dispersion point") reduces the computational complexity of traditional clustering algorithms from nm times to a fixed 4 cross-cluster representative point distance calculations, significantly reducing computational power requirements and operating costs, and substantially improving clustering efficiency to meet the needs of large-scale resource clusters. Simultaneously, using the median as the inter-cluster distance metric, combined with the comprehensive characterization of cluster features by the dual-representative points, it effectively filters out off-samples, data biases, and noise interference. Even in scenarios where resource data comes from multiple operating entities, has varying data quality, or has a small sample size, it still ensures the stability and accuracy of clustering results, truly reflecting resource response characteristics. Furthermore, the method considers both temporal characteristics and dynamic iteration capabilities, adapting in real-time to the dynamic changes in storage, charging, and emptying resources, providing a scientific basis for subsequent refined scheduling and optimized allocation of resources, ultimately achieving a dual improvement in the utilization efficiency and management level of storage, charging, and emptying resources in building industrial parks.

[0137] like Figure 2 As shown, a hierarchical and dynamic clustering system for storage, charging, and emptying resources in a building park includes:

[0138] The feature vector module is used to classify building parks into different levels based on their geographical location and the type of energy storage, charging, and air conditioning resources, and to collect the actual operating data of the energy storage, charging, and air conditioning equipment at each level to construct the corresponding three-dimensional feature vector; the energy storage, charging, and air conditioning equipment can be energy storage equipment, charging piles, or air conditioners.

[0139] The distance calculation module is used to combine the three-dimensional characteristic vectors of each storage, charging and emptying device with the interactive scene, and calculate the distance between each pair of different storage, charging and emptying devices at each level through an adaptive distance weight allocation algorithm and aggregate them to generate an initial cluster.

[0140] The clustering module is used to calculate the inter-cluster distance of each initial cluster based on the cluster center and the maximum dispersion point of each initial cluster, and to perform clustering iteration to generate resource aggregation results.

[0141] The distance calculation module includes:

[0142] The weight preset unit is used to set the basic weights of different types of storage, charging and emptying devices as cluster points at each level; the basic weights include power basic weight, response time basic weight and cost basic weight.

[0143] The first coefficient calculation unit is used to set the adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and to calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment.

[0144] The second coefficient calculation unit is used to calculate the scene correction coefficient of each level based on the three-dimensional characteristic vectors and interactive scenes of each storage, filling and emptying device through the scene judgment function.

[0145] The weight update unit is used to update the basic weights of each level based on the scene correction coefficient and state correction coefficient of each level, and generate the corresponding final weights.

[0146] The distance calculation unit is used to calculate and aggregate the distances between pairs of different clusters at the same level based on the final weights of each level, thus obtaining the corresponding initial clusters.

[0147] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0148] Example 2:

[0149] This embodiment is a simulation based on Embodiment 1. Taking a dynamic hierarchical clustering of 20 energy storage devices, 100 charging piles, and 200 air conditioners in a building complex as an example, the energy storage devices are in a frequency regulation scenario. The SOC of the 20 energy storage devices ranges from 0.1 to 0.9, with adjustment coefficients for power, response time, and cost of 0.8, 0.4, and -0.6, respectively. The weighting benchmarks for power, response time, and cost are 0.6, 0.8, and 1.3, respectively. The charging piles are in a peak shaving scenario, with adjustment coefficients for response time and cost of 0.6 and 0.8, respectively. The weighting benchmarks for response time and cost are 0.7 and 0.6, respectively. The user expects the total time for the vehicle to connect to the charging pile to range from [2h, 8h]. The user's allowed adjustment window for the charging period is 0.1 to 0.7 times the user expects the total time for the vehicle to connect to the charging pile to reach the charging pile, and the behavioral response level ranges from [0, 1]. The air conditioner is in an economic scheduling scenario, with power and cost adjustment coefficients of 0.6 and 0.4 respectively, weighting benchmarks of power and cost of 0.7 and 0.8 respectively, indoor temperature range of [22°C, 30°C], user-set temperature range of [24°C, 35°C], maximum temperature deviation of 5 degrees Celsius, and user comfort requirement range of [0.3, 1].

[0150] In this embodiment, the dynamic clustering period is 15 minutes, and clustering iterations are performed with a time step of 5 minutes. The final simulation results should include a tree diagram of the aggregation process for each resource type, a cluster diagram of each resource point in three-dimensional space, and a cross-time period analysis diagram of the affiliation of each resource cluster.

[0151] In this embodiment, the power range of the energy storage device is [-200kW, 200kW], the response duration range is [0h, 6h], and the price range is [0.4 yuan / kW, 0.6 yuan / kW]. The power range of the charging pile is [-100kW, 100kW], the response duration range is [0h, 4h], and the price range is [0.3 yuan / kW, 0.4 yuan / kW]. The power range of the air conditioner is [0kW, 100kW], the response duration range is [0h, 3h], and the price range is [0.1 yuan / kW, 0.3 yuan / kW].

[0152] Clustering results of energy storage devices in frequency regulation scenarios are as follows: Figure 3 and Figure 4As shown, the process and effect of dynamic hierarchical clustering are clearly presented. From the perspective of time steps, at three step sizes of 5 / 10 / 15 minutes, the clustering dendrogram, based on weighted distance, orderly displays the hierarchical aggregation process of agglomerative hierarchical clustering, reflecting the gradient matching of similarity between clusters. The device points of the three clusters (red / green / blue) have strong spatial aggregation and clear boundaries, reflecting that the improved algorithm can accurately distinguish energy storage devices with different characteristics. The cluster size is dynamically adjusted at each step size (e.g., cluster 2 contains 8 devices at 5 minutes and increases to 11 devices at 10 minutes), adapting to the real-time changes in the operating status of the devices. It intuitively reflects that the cluster affiliation of 20 energy storage devices is updated with the step size (e.g., devices 3, 7, etc. switch clusters at different times), matching the time-varying characteristics of their SOC and other operating statuses.

[0153] Clustering results of charging piles in peak shaving scenarios are as follows Figure 5 and Figure 6 As shown, at step sizes of 5 / 10 / 15 minutes, the clustering dendrogram presents the hierarchical aggregation process of agglomerative clustering using weighted distance, reflecting the gradient matching of similarity between clusters. The 100 charging piles in the three clusters (red / green / blue) have strong spatial aggregation and clear boundaries, reflecting that the algorithm can accurately distinguish devices with different charging durations and behavioral response levels. It shows the dynamic adjustment of cluster size at each step size (e.g., cluster 3 contains 53 units at 5 minutes and 48 units at 10 minutes), which matches the real-time changes in the user's charging time window. The cross-time period affiliation graph shows that the cluster affiliation of charging piles is updated with the step size, matching the time-varying characteristics of user charging behavior.

[0154] Clustering results of air conditioners in economic dispatch scenarios are as follows Figure 7 and Figure 8 As shown, with step sizes of 5 / 10 / 15 minutes, the 200 air conditioner points in the 6 clusters are highly concentrated in the space with clear boundaries, accurately distinguishing devices with different temperature deviations and user comfort requirements. The cluster size is dynamically adjusted at each step size (e.g., 5 minutes for cluster 6 with 51 units, 10 minutes for 54 units, and 15 minutes for a more balanced cluster size), which matches the real-time changes in air conditioning temperature and user comfort needs, reflecting the time-varying characteristics of air conditioner cluster affiliation as the step size is updated to match their operating status.

[0155] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0156] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for hierarchical classification and dynamic clustering of storage, replenishment, and empty resources in a building park, characterized in that, include: Based on the geographical location and energy storage / charging / air conditioning resource types of the building park, the building park is classified into layers, and the actual operation data of the energy storage / charging / air conditioning equipment at each layer is collected to construct the corresponding three-dimensional characteristic vector; the energy storage / charging / air conditioning equipment is energy storage equipment, charging pile, or air conditioning. By combining the three-dimensional characteristic vectors of each storage, charging and emptying device and the interactive scenario, the distance between each pair of different storage, charging and emptying devices at each level is calculated and aggregated using an adaptive distance weight allocation algorithm to generate an initial cluster; Based on the cluster centers and maximum dispersion points of each initial cluster, the inter-cluster distances of each initial cluster are calculated using an improved agglomerative hierarchical clustering algorithm, and clustering iterations are performed to generate resource aggregation results.

2. The method for hierarchical classification and dynamic clustering of storage, replenishment, and empty space resources in a building park according to claim 1, characterized in that, The generation of the initial cluster includes: Each storage, charging, and emptying device at each level is used as a cluster point, and basic weights are set for different types of storage, charging, and emptying devices; the basic weights include power basic weight, response time basic weight, and cost basic weight; Set adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment; Based on the three-dimensional characteristic vectors and interactive scenarios of each storage, charging and emptying device, the scenario correction coefficient for each level is calculated through the scenario judgment function; Based on the scene correction coefficient and state correction coefficient of each level, update the basic weights of each level and generate the corresponding final weights. Based on the final weights of each level, the distance between any two different clusters at the same level is calculated and aggregated to obtain the corresponding initial clusters.

3. The method for hierarchical classification and dynamic clustering of storage, replenishment, and empty space resources in a building park according to claim 2, characterized in that, The calculation of the state correction coefficient for each level includes: When the When the energy storage and charging equipment at each level is an energy storage device, the calculation is based on the adjustment coefficient and weighting benchmark of the energy storage device and the corresponding state of charge of each energy storage device. State correction coefficients for each level of energy storage equipment; When the When the energy storage and charging equipment at each level is a charging pile, calculate the charging-dwell time ratio and behavioral response level of each charging pile; based on the charging-dwell time ratio, behavioral response level, and corresponding adjustment coefficients and weighting benchmarks of each charging pile, calculate the first... Status correction coefficients for each charging pile at each level; When the When the storage and charging equipment at each level is an air conditioner, the indoor temperature and user-set temperature of each air conditioner are collected, and the temperature deviation value of each air conditioner is calculated based on the set maximum temperature deviation; based on user comfort requirements, the temperature deviation value of each air conditioner, the adjustment coefficient, and the weighting benchmark, the first... The status correction coefficients for each air conditioner at each level.

4. The method for hierarchical classification and dynamic clustering of storage, replenishment, and empty space resources in a building park according to claim 2, characterized in that, The calculation of the scene correction coefficient for each level using the scene judgment function includes: Set frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold to obtain the current electricity price and grid frequency deviation at each level; Collect historical operational data of the building park and use machine learning or deep learning algorithms to make predictions to obtain the load forecast value at the current moment; Based on the frequency regulation action threshold, peak-valley difference threshold, and electricity price fluctuation threshold, the load forecast, electricity price, and grid frequency deviation at each level are judged by the scenario judgment function to determine the interaction scenario at each level. Based on the interactive scenarios at each level, corresponding scenario correction coefficients are generated.

5. The method for hierarchical classification and dynamic clustering of storage, replenishment, and empty space resources in a building park according to claim 4, characterized in that, The formula corresponding to the scenario judgment function is: ； in, Represents the maximum value function. Describes the minimum value function. This indicates the current system load status of the building complex. This represents the scene judgment function. , These represent the power grid frequency deviation and the frequency regulation action threshold, respectively. Represents absolute value. , Representing time respectively The load forecast values ​​for the building park and the set peak-valley difference threshold, , Representing time respectively Electricity price and electricity price fluctuation threshold, , , These represent frequency modulation scenarios, peak shaving scenarios, and economic dispatch scenarios, respectively.

6. The method for hierarchical classification and dynamic clustering of storage, charging, and empty resources in a building park according to claim 1, characterized in that, The generated resource aggregation results include: Select any level and use the corresponding initial cluster as the starting point for iteration; Determine the cluster center of each initial cluster and use it as the first representative point. Calculate the distance between different points in each initial cluster and the cluster center. Select the point with the largest dispersion in each initial cluster and use it as the second representative point. Calculate the distances between the first and second representative points in each pair of different initial clusters and sort them from largest to smallest. Select the distance that is the median as the inter-cluster distance between the two corresponding initial clusters. Iterate through all inter-cluster distances, select the pair of clusters with the smallest distance and merge them to generate a new cluster and iterate until the preset conditions are met to obtain the resource aggregation result of the selected level; Reselect any level and iterate until you obtain the resource aggregation results for all levels.

7. A hierarchical and dynamic clustering system for storage, replenishment, and empty space resources in a building park, characterized in that: include: The feature vector module is used to classify building parks into different levels based on their geographical location and the type of energy storage, charging, and air conditioning resources, and to collect the actual operating data of the energy storage, charging, and air conditioning equipment at each level to construct the corresponding three-dimensional feature vector; the energy storage, charging, and air conditioning equipment can be energy storage equipment, charging piles, or air conditioners. The distance calculation module is used to combine the three-dimensional characteristic vectors of each storage, charging and emptying device with the interactive scene, and calculate the distance between each pair of different storage, charging and emptying devices at each level through an adaptive distance weight allocation algorithm and aggregate them to generate an initial cluster. The clustering module is used to calculate the inter-cluster distance of each initial cluster based on the cluster center and the maximum dispersion point of each initial cluster, and to perform clustering iteration to generate resource aggregation results.

8. A hierarchical classification and dynamic clustering system for storage, replenishment, and emptying resources in a building park, as described in claim 7, is characterized in that... The distance calculation module includes: The weight preset unit is used to set the basic weights of different types of storage, charging and emptying equipment as cluster points for each level; the basic weights include power basic weight, response time basic weight and cost basic weight. The first coefficient calculation unit is used to set the adjustment coefficients and weight benchmarks for different types of storage, filling and emptying equipment, and to calculate the state correction coefficients for each level based on the three-dimensional characteristic vectors of each storage, filling and emptying equipment. The second coefficient calculation unit is used to calculate the scene correction coefficient of each level based on the three-dimensional characteristic vectors and interactive scenes of each storage, filling and emptying device through the scene judgment function. The weight update unit is used to update the basic weights of each level based on the scene correction coefficient and state correction coefficient of each level, and generate the corresponding final weights. The distance calculation unit is used to calculate and aggregate the distances between pairs of different clusters at the same level based on the final weights of each level, thus obtaining the corresponding initial clusters.

Images