Photovoltaic power generation absorption system based on distributed test unit

Abstract

The invention relates to a photovoltaic power generation consumption system based on a distributed test unit, the photovoltaic power generation consumption system is deployed in a micro-grid architecture, the micro-grid architecture comprises a power utilization node, a photovoltaic power generation unit and external commercial power, the photovoltaic power generation unit is provided with a photovoltaic inverter controller, the power utilization node is provided with an energy storage unit, and the external commercial power is connected with the photovoltaic inverter controller. The distributed test unit cluster collects power utilization load data in real time, the photovoltaic monitoring unit cluster obtains power generation power data, the power utilization rule analysis unit generates a power utilization rule portrait, and the power supply matching unit dynamically establishes a power supply relation to realize efficient matching of photovoltaic power generation and power utilization demands. The method has the advantages of improving the local consumption efficiency of photovoltaic energy and guaranteeing the power utilization quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of photovoltaic power generation technology, and more particularly, to a photovoltaic power generation consumption system based on a distributed test unit. BACKGROUND

[0002] In the field of renewable energy power generation technology, photovoltaic power generation systems are widely used due to their clean and sustainable characteristics. Grid-connected operation and power consumption are key links to ensure the stability and efficiency of power systems. Among them, load prediction and dynamic matching of power generation power constitute the core technical problem of system operation.

[0003] Among them, the distributed photovoltaic power consumption system aims to realize efficient matching between photovoltaic power generation power and electricity load through prediction and scheduling means. This system usually involves monitoring, analyzing and optimizing control of multiple distributed photovoltaic units and electricity loads in the region to improve the local consumption rate of photovoltaic energy and reduce the impact on the upper power grid.

[0004] Existing technologies mostly use centralized prediction models or single load regulation strategies, which are difficult to effectively cope with the spatial and temporal differences of electricity consumption rules in distributed scenarios. Traditional methods have significantly reduced prediction accuracy under conditions of severe photovoltaic output power fluctuations or complex and variable load demands, and lack dynamic identification and configuration mechanisms for electricity end priority. For example, a photovoltaic power generation load prediction method based on prediction accuracy (CN115526414B). This patent analyzes historical photovoltaic data and natural climate data to construct a short-term prediction model for traditional feeder load and photovoltaic power generation load, and combines the prediction results of both to improve prediction accuracy. However, this technical solution is mainly aimed at centralized load prediction models. For example, a photovoltaic load prediction method and device based on artificial intelligence (CN119051018B). This patent uses a generative adversarial network to expand the training data set, and combines a feature extraction model and a classifier model to realize load prediction interval output. This method to some extent solves the problem of insufficient training data and improves the generalization ability of the prediction model. However, this technical solution focuses on improving the accuracy of load prediction and fails to delve into the priority division and dynamic matching strategy of different electricity ends in distributed photovoltaic systems. At the same time, this method does not involve the specific mechanism of energy storage intervention and external power grid collaborative optimization, making it difficult to cope with fluctuations in power supply demand in complex scenarios, which may affect the overall power supply stability and economy of the system.

[0005] The main problem is that the consumption capacity of photovoltaic renewable resources is insufficient, and in most cases, photovoltaic power needs to be transmitted to the energy storage station of the power grid through grid connection for storage, and then distributed to each user node that needs it, which causes a problem that the line loss of photovoltaic energy is large during transmission, and the storage energy dispatching needs to occupy resources, and although some technologies disclose the idea of localizing the use of photovoltaic energy, it is generally limited to the use of photovoltaic power generation units of users for their own use, and if it involves cross-user scheduling, there will be a problem that the variability of user load and the supply capacity of photovoltaic power generation units do not match, resulting in the quality of user power being affected. Therefore, how to ensure the quality of power in the case of full scheduling of photovoltaic resources is a research topic. SUMMARY

[0006] Therefore, the present application aims to provide a photovoltaic power consumption system based on a distributed test unit.

[0007] In order to solve the above technical problems, the technical scheme of the present application is:

[0008] A photovoltaic power consumption system based on a distributed test unit is deployed in a micro-grid architecture, the micro-grid architecture includes a power consumption node, a photovoltaic power generation unit, and an external power supply, the photovoltaic power generation unit is configured with a photovoltaic inverter controller, and the power consumption node is configured with an energy storage unit, characterized in that it comprises:

[0009] A distributed test unit cluster is deployed in each power consumption node for real-time collection and uploading of power consumption load data;

[0010] A photovoltaic monitoring unit cluster is deployed in the photovoltaic power generation unit for obtaining output power data and environmental monitoring data of the photovoltaic power generation unit;

[0011] A power supply distributor is coupled to the photovoltaic power generation unit on one side and to the power consumption node on the other side, and the power supply distributor is used to connect the power consumption node to the power supply loop of the photovoltaic power generation unit according to the matching relationship between the photovoltaic power generation unit and the power consumption node;

[0012] A power consumption rule analysis unit is configured with a rule identification strategy, which is used for mode identification and feature extraction of the collected power consumption load data under multiple time scales to generate a power consumption rule portrait;

[0013] The power supply matching unit is used to establish a power supply matching relationship based on the power consumption law portrait, so that each power generation node is matched with at least one photovoltaic power generation unit, and each photovoltaic power generation unit is configured with a dynamic power supply strategy corresponding to each power generation node. The existing scheme relies on centralized prediction and unified scheduling, which cannot adapt to dynamic changes in distributed scenarios. The present scheme realizes real-time synchronization of data through a distributed monitoring network, and improves the power consumption law identification accuracy by combining multi-time scale analysis. The power supply distributor directly establishes the dynamic connection between the power generation unit and the power consumption node, reducing the line loss caused by traditional grid-connected transmission. The dynamic power supply strategy adjusts the power supply topology according to the real-time supply and demand state, optimizes the utilization of energy storage resources while ensuring power quality, and avoids excessive dependence on external power. It can reduce the line loss in the process of photovoltaic energy transmission, reduce the occupation of energy storage resources, and improve the local consumption rate through dynamic matching of supply and demand. The system dynamically adjusts the power supply topology according to real-time monitoring data and power consumption law portrait, maintains the balance between supply and demand when photovoltaic output fluctuates or load suddenly changes, and avoids the problem of power quality decline caused by traditional cross-user scheduling.

[0014] Further, the law identification strategy includes:

[0015] Step A1, receiving the power consumption load time series data of the target power consumption node;

[0016] Step A2, preprocessing the time series data, including missing value interpolation and outlier removal;

[0017] Step A3, segmenting the power consumption load time series data by multi-time label;

[0018] Step A4, matching each segment by a pre-set typical load characteristic model to obtain the optimal typical load curve of each segment, and generating a typical matching value for each segment;

[0019] Step A5, fitting each segment by a pre-set dynamic law fitting sub-strategy and generating a corresponding law fitting value;

[0020] Step A6, generating a law matching vector with the classic matching value and the law fitting value as components, and obtaining the corresponding power consumption law portrait from the pre-set law portrait library according to the law matching vector. Traditional methods use single time scale analysis of load data, which is difficult to cope with the spatiotemporal differences of power consumption law in distributed scenarios. The present scheme solves the problem of insufficient prediction accuracy caused by the spatiotemporal differences of power consumption load in distributed photovoltaic systems. Through the multi-time label segmentation mechanism, the load data is finely processed, the typical load characteristic model matching ensures the accurate extraction of common laws, and the dynamic law fitting sub-strategy enhances the modeling ability of atypical fluctuations. The finally generated power consumption law portrait provides high-precision data support for the establishment of power supply matching relationship, effectively improving the dynamic matching efficiency of photovoltaic power generation and power consumption load.

[0021] Further, the rule identification strategy further includes a model construction sub-strategy, and the model construction sub-strategy includes:

[0022] Step A41, a density clustering-based unsupervised learning algorithm is used to identify typical load curve clusters of different power consumption nodes at daily, weekly and monthly time scales, and corresponding multi-time labels are generated;

[0023] Step A42, according to the multi-time labels, the preset random test samples are split according to the typical load curve clusters to generate explanation information;

[0024] Step A43, whether the explanation information meets the preset explanation condition is evaluated, if yes, step A44 is entered, and if not, cluster adjustment instructions are generated according to the explanation information to adjust the corresponding typical load curve clusters, and step A42 is returned;

[0025] Step A44, according to each typical load curve cluster, a corresponding typical load curve, a matching feature and a bias influence mapping function are generated to generate the typical load feature model. The conventional method usually extracts the load mode by using a fixed time window, and cannot adaptively adjust the model structure. The existing clustering algorithm is usually only for a single time scale, and lacks multi-dimensional correlation analysis capability. The present scheme realizes model iterative optimization through a dynamic evaluation mechanism, and constructs cross-scale correlation in combination with multi-time labels, effectively solving the time and space difference problem of power consumption mode in a distributed scene. The present application realizes adaptive construction of the typical load feature model, and improves the time and space matching precision of power consumption rule identification. Through the multi-time label and the dynamic adjustment mechanism, the generalization ability of the model to different power consumption scenes is enhanced. The iterative optimization process driven by the explanation information ensures the consistency of the model and the actual power consumption behavior, and provides a reliable load feature analysis basis for subsequent power supply matching.

[0026] Further, the typical load characteristic model generates a typical difference curve by differencing the segmented and typical load curve of the time series data of the power load, calculates a deviation reduction value through a deviation influence mapping function, extracts a corresponding matching feature in the segment according to the matching feature to obtain a corresponding matching gain value, and obtains a corresponding typical matching value by inputting the deviation reduction value and the matching gain value into a preset matching evaluation algorithm. By introducing a two-dimensional evaluation system of matching gain value and deviation reduction value, the quantification ability of the overall deviation is retained, and the weight distribution of the core feature matching is strengthened. By dynamically adjusting the deviation influence mapping function and the matching feature extraction rule, the load fluctuation characteristics in different scenarios can be adapted, and the defects of the fixed threshold matching mechanism are overcome. It can effectively distinguish between systematic deviation and random noise in load fluctuation, improve the matching accuracy of the typical load curve. In the dynamic matching process of the photovoltaic power generation unit and the power consumption node, the technical scheme can accurately identify the core matching parameters of the load characteristics, reduce the interference of abnormal fluctuations on the power supply strategy, and thus improve the dynamic response accuracy and power supply stability of the photovoltaic power generation consumption system.

[0027] Further, the dynamic rule fitting sub-strategy includes:

[0028] Step A51, call the qualified reference load period through the multi-time label;

[0029] Step A52, establish the period association between segments by the obtained reference load period, so that each segment has at least one associated segment to construct a corresponding period association group, and the period associations of the segments of different period association groups are different;

[0030] Step A53, calculate the theoretical waveform of each segment through the period association and obtain the period deviation waveform by differencing the actual waveform;

[0031] Step A54, calculate the period deviation value of each period association group through the preset period rule evaluation algorithm;

[0032] Step A55, select the period correlation group with the smallest period deviation value, extract the corresponding period matching density, the period matching density is the number of period correlations in the period correlation group divided by the number of segments, and determine the rule fitting value according to the period matching density and the period deviation value. Based on the multi-time label, the reference load period with similar time attribute to the current load segment is screened out. By establishing the period correlation relationship between the segments, the period correlation group with different correlation characteristics is formed. For each correlation group, the deviation waveform of the theoretical load waveform and the actual waveform is calculated, and the period rule evaluation algorithm is used to quantify the deviation degree of each group. After selecting the correlation group with the smallest deviation, the period matching density and the deviation value of the group are combined for weighted calculation to generate a numerical index reflecting the quality of dynamic rule fitting. At the same time, the existing technology mainly adopts fixed period template matching, while the scheme dynamically constructs the correlation group and calculates the comprehensive index of matching density and deviation value, which significantly improves the adaptability and accuracy of rule fitting.

[0033] Further, the power supply matching unit is configured with a power supply matching strategy for determining the power supply matching relationship, and the power supply matching strategy includes:

[0034] Step B1, generating a power generation characteristic image according to the historical power generation data and power generation characteristic parameters of the power generation unit;

[0035] Step B2, marking the priority of the power consumption rule image and the power generation characteristic image according to the preset grading standard;

[0036] Step B3, matching the power consumption rule image and the power generation characteristic image located in the same priority to divide into corresponding consumption matching groups;

[0037] Step B4, obtaining the average power consumption load of each power consumption node in the consumption matching group and the average power generation supply of each power generation unit, and establishing the power supply matching relationship between the power consumption node and the power generation unit according to the loss target constraint, so that each power consumption node matches at least one power generation unit and meets the preset power supply requirement constraint. The power supply requirement constraint is that the power supply and demand difference of the power consumption node and the power generation unit with the power supply matching relationship is within the preset reference supply and demand range, and the reference supply and demand range corresponding to each priority is different. The dynamic change rule of the power consumption load in the distributed photovoltaic system is difficult to accurately fit. By establishing the period correlation relationship under multiple time scales, the potential periodicity characteristics of the load change are effectively identified, and the rule misjudgment risk caused by single time scale analysis is reduced. Combined with the comprehensive calculation mechanism of period matching density and deviation value, accurate rule fitting quantitative index is provided for the power supply matching strategy, thereby improving the dynamic matching accuracy of photovoltaic power generation and power consumption load and ensuring the stable operation of the power supply system.

[0038] Further, the loss target constraint is configured with a loss target function that minimizes the loss target value of the consumption matching group after establishing the power supply matching relationship, and the loss target function calculates the loss target value based on a transmission loss value, a quality matching loss value, and a storage loss value. The transmission loss value reflects the loss in the transmission process, the quality matching loss value reflects the loss caused by the mismatch of power supply quality, and the storage loss value reflects the loss corresponding to the storage demand caused by excessive power supply. The problem of low efficiency of supply and demand dynamic matching in the distributed photovoltaic system is effectively solved. Through the quantitative matching of the power generation characteristic image and the power consumption rule image, the precise adaptation of the supply and demand characteristics is realized, and the decline of power quality caused by cross-user scheduling is avoided. The introduction of hierarchical standards and dynamic benchmark range makes high-priority nodes obtain more stringent power supply quality guarantee, and low-priority nodes fully utilize the power supply flexibility. The division of the consumption matching group reduces the system matching complexity and improves the real-time scheduling efficiency. The optimization calculation of the loss target constraint ensures the balance between power supply stability and resource utilization, and reduces the storage demand and transmission loss caused by supply and demand mismatch.

[0039] Further, the dynamic power supply strategy includes:

[0040] Step C1, generating a corresponding power supply topology level according to the real-time acquired power data, environmental monitoring data and power consumption load data through a preset topology evaluation algorithm, and configuring the power supply topology of the power consumption node according to the power supply topology level. Under different priorities, the mapping relationship between the power supply topology level and the power supply topology is different;

[0041] Step C2, obtaining the storage response constraint according to the power supply topology, judging the state data of the corresponding storage unit through the storage response constraint, and executing the corresponding execution sub-strategy according to the state data. Different dynamic power supply strategies correspond to different storage response constraints. The response lag and mode conflict problems existing in the dynamic adjustment of the power supply topology are solved. In the photovoltaic output sudden drop scenario, the system quickly identifies the power supply risk through real-time scoring and automatically switches to the power supply topology containing the backup unit; in the case of load surge, the optimal power supply mode is selected according to the real-time state of the storage unit, avoiding power interruption caused by insufficient storage capacity. This scheme maintains the power quality while improving the local consumption rate of photovoltaic energy and reducing the dependence on external municipal power.

[0042] Further, the power supply topology includes a first power supply topology, a second power supply topology, a third power supply topology, a fourth power supply topology and a fifth power supply topology, the first power supply topology is that only the photovoltaic unit of the matching group is connected to the power supply loop of the power supply node, the second power supply topology includes that the photovoltaic unit of the matching group and the standby photovoltaic unit are connected to the power supply loop of the power supply node, the third power supply topology includes that the photovoltaic unit of the matching group and the standby energy storage unit are connected to the power supply loop of the power supply node, the fourth power supply topology includes that the photovoltaic unit of the matching group and the external power supply are connected to the power supply loop, and the fifth power supply topology includes that the external power supply is connected to the power supply loop. Through the hierarchical configuration of the five topologies, a gradual switching path from pure photovoltaic power supply to power supply is realized, and the photovoltaic energy is preferentially used while the power supply continuity is ensured. For example, when the light intensity suddenly drops but the energy storage unit still has capacity, the traditional method may directly enable the power supply, while the present scheme preferentially enables the third topology to use the energy storage to maintain the power supply, thereby reducing the proportion of power supply use.

[0043] Further, the execution sub-strategy includes executing an energy storage load mode, an energy storage mode, a load mode and an energy storage cooperative supply mode, the energy storage load mode is that the photovoltaic unit simultaneously supplies power to the energy storage unit and the power consumption node, the energy storage mode is that the photovoltaic unit only supplies power to the energy storage unit, the load mode is that the photovoltaic unit only supplies power to the power consumption node, and the energy storage cooperative supply mode is that the energy storage unit and the photovoltaic unit simultaneously supply power to the power consumption node. The present application can quickly switch the power supply mode when the photovoltaic output fluctuates, for example, the energy storage cooperative supply mode is started immediately when the photovoltaic output suddenly drops in rainy weather, so as to avoid voltage drop on the load side; the energy storage mode is automatically switched to when the photovoltaic output is at the peak at noon and the load demand is low, thereby reducing energy waste. Through the dynamic matching of the energy storage state and the power supply loop, the present application realizes the dual optimization of power supply quality and energy utilization efficiency.

[0044] The technical effects of the present application mainly embody in the following aspects: through the distributed test unit cluster, the power consumption load data is collected in real time, the power generation power data is obtained by combining the photovoltaic monitoring unit cluster, the power consumption rule portrait is generated by using the power consumption rule analysis unit, and the power supply relationship is dynamically established by using the power supply matching unit, so as to realize the efficient matching of photovoltaic power generation and power consumption demand, and has the advantages of improving the local consumption efficiency of photovoltaic energy and guaranteeing the power consumption quality. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The architecture principle diagram of the photovoltaic power generation consumption system based on the distributed test unit of the present application;

[0046] Figure 2 The flow chart of the rule identification strategy of the present application;

[0047] Figure 3A flow chart of the model construction sub-strategy of the present application;

[0048] Figure 4 A flow chart of the dynamic fitting sub-strategy of the present application;

[0049] Figure 5 A flow chart of the dynamic rule fitting sub-strategy of the present application. DETAILED DESCRIPTION

[0050] The specific embodiments of the present application are further described in detail below with reference to the accompanying drawings, so that the technical solutions of the present application are easier to understand and master.

[0051] A photovoltaic power generation consumption system based on a distributed test unit is deployed in a micro-grid architecture, the micro-grid architecture includes a power consumption node, a photovoltaic power generation unit, and an external power supply, the photovoltaic power generation unit is configured with a photovoltaic inverter controller, and the power consumption node is configured with an energy storage unit, characterized in that it comprises:

[0052] A distributed test unit cluster is deployed in each power consumption node for real-time acquisition and uploading of power consumption load data;

[0053] A photovoltaic monitoring unit cluster is deployed in the photovoltaic power generation unit for implementation of obtaining output power data of the photovoltaic power generation unit and environmental monitoring data;

[0054] A power supply distributor is coupled to the photovoltaic power generation unit on one side and to the power consumption node on the other side, and the power supply distributor is used to connect the power consumption node to the power supply loop of the photovoltaic power generation unit according to the matching relationship between the photovoltaic power generation unit and the power consumption node; a power consumption rule distribution analysis unit is configured in the photovoltaic inverter controller, and correspondingly, a power supply matching unit is also provided in the photovoltaic inverter controller, and through feedback of corresponding signals by the EMS and user terminal acquisition unit and the DD battery controller, the inverter can obtain data of the corresponding system in real time during operation, so that a matching relationship between the photovoltaic unit and the power consumption node is established at the same time, and a digital communication matching relationship is also established, so as to complete the execution of specific instructions.

[0055] A power consumption rule analysis unit is configured with a rule identification strategy, which is used for mode recognition and feature extraction of the acquired power consumption load data under multiple time scales to generate a power consumption rule image;

[0056] The rule identification strategy comprises:

[0057] Step A1: Receive the time-series data of the target electricity consumption node. The "target electricity consumption node" refers to a terminal node in the distributed microgrid with independent electricity demand and configured with energy storage units. Each node is assigned a unique identifier to associate with the distributed test unit. Node types include residential user nodes, commercial electricity consumption nodes, and industrial auxiliary electricity consumption nodes. The differences in electricity consumption characteristics among different node types are differentiated and adapted through subsequent analysis strategies. The time-series data of the electricity load refers to the power change dataset continuously collected by the distributed test unit at a preset sampling frequency. The sampling frequency is dynamically configured according to the node type, with residential user nodes sampling every 15 minutes by default and commercial and industrial auxiliary electricity consumption nodes every 5 minutes by default. Data dimensions include the collection timestamp, instantaneous active power value, and cumulative energy consumption value. The data format adopts the JSON standard format to ensure cross-unit transmission compatibility. During data reception, the unit will perform a preliminary verification of data integrity. If the data loss rate of a single batch exceeds 5%, a retransmission command will be sent to the distributed test unit to ensure the validity of the original data.

[0058] Step A2: Preprocess the time series data, including missing value imputation and outlier removal. Missing values ​​are caused by communication interruptions in the distributed test unit, temporary equipment failures, etc. Different methods are used depending on the number of missing consecutive sampling points: when the number of missing points is ≤3, a weighted sliding window linear interpolation method is used. Let the timestamp of the missing point be t, and the adjacent valid data points be t1 / power P1 and t2 / power P2, respectively. The sliding window size is N=5, including the 2 valid points before t1 and the 2 valid points after t2. First, calculate the time weight ω of each valid point within the window. i =1 / |tt i |,t i For the valid timestamp, calculate the interpolation power P for the missing points:

[0059]

[0060] Where i∈[1,N], N is the sliding window size, ωi is the time weight of the i-th valid point, and P i Let P be the power value of the i-th valid point within the window. This method uses time weighting to increase the influence of neighboring data on missing values, ensuring that the interpolation closely matches the actual load trend. When the number of missing points is greater than 3, the historical similar day trend completion method is used. First, historical dates with the same date type as the current missing period and a meteorological condition similarity of ≥85% are selected as similar days. Load data for the corresponding period of the similar days are extracted, and then adjusted by linear scaling as the interpolation value. The power P is scaled as follows:

[0061] P = K × P

[0062] Wherein, K is the scaling factor, P similar day is the power value of the similar day corresponding period, the method is suitable for long time missing scene, and the rationality of the interpolation data is guaranteed through meteorological and date matching.

[0063] The abnormal value refers to the power value deviating from the normal range due to equipment failure and sudden power consumption. The processing flow is divided into three steps: first, the average power Paverage and the standard deviation s of the 1-hour period in which a sampling point is located are calculated. If the power P of the point i satisfies |Pi-Paverage|>3xs, it is marked as an abnormal candidate point; then secondary verification is performed, and the power change rate AP=|Pi-Pi-1| of the candidate point and the previous and subsequent two sampling points is calculated i i-1 i-1 If AP exceeds the node type threshold and there is no corresponding equipment start-stop record, the abnormality is confirmed; finally, abnormal replacement is performed: the weighted sliding window linear interpolation method consistent with the missing value interpolation is used to replace the abnormal value, ensuring the continuity of the time series.

[0064] Step A3, the electricity load time series data is segmented by multi-time label; the "multi-time label" is a three-dimensional time identification set designed based on the periodic characteristics of electricity load, including day time label, week time label, and month time label, which is used to divide the load period with similar electricity consumption rules. The day time label is divided according to the load peak and valley in a day, including early peak segment, flat peak segment, late peak segment (18:00-24:00), and valley segment; the week time label is divided according to the weekly electricity consumption behavior, including weekday label and weekend label; the month time label is divided according to the seasonal electricity demand, including winter label, spring label, summer label, and autumn label. The multi-time label assignment rule is to match the "day-week-month" three-dimensional label to each sampling point timestamp in turn, for example, the label of "202X year 7 month 15 day (Monday) 19:30" is "late peak segment-weekday-summer". The segmentation process is progressive according to three levels: the first level is divided into different seasonal segments according to the month time label; the second level is divided into weekday and weekend sub-segments according to the week time label within the seasonal segment; the third level is divided into the final segments according to the day time label within the weekday / weekend sub-segment, which requires that all sampling points in each segment have completely consistent labels and the time length is greater than or equal to 2 hours. If the time length corresponding to a certain label combination is less than 2 hours, it is merged with the adjacent segment with consistent week and month labels, and the day label with the highest proportion is taken as the day time label of the merged segment.

[0065] ​​Step A4, match each segment by preset typical load characteristic model to obtain the optimal typical load curve of each segment, and generate the typical matching value of each segment; for each segment, first extract its multi-time label, call the candidate typical load curve corresponding to the label from the model, and then filter the optimal curve through characteristic similarity calculation: extract matching characteristics: include load peak value Pmax, valley value Pmin, average value Pavg, fluctuation coefficient Cv=σ / Pavg, σ is the standard deviation of the segment, peak-valley difference ΔP, 5 indexes; first, the relative error needs to be calculated: for each characteristic j, calculate the relative error εj=|Xjactual-Xjcandidatedjcandidateof the actual value and the characteristic value of the candidate curve.

[0066] Calculate total error: set feature weight wj according to node type, the value of residential node wjis w1=0.2, w2=0.1, w3=0.3, w4=0.2, w5=0.2; the value of commercial node wjis w1=0.3, w2=0.1, w3=0.2, w4=0.2, w5=0.2; the value of industrial auxiliary node wjis w1=0.25, w2=0.15, w3=0.25, w4=0.15, w5=0.2, calculate the total error E by the following formula:

[0067]

[0068] Where, j∈[1,5] corresponds to 5 matching characteristics, w j is the weight of the jth characteristic, ε j is the relative error of the jth characteristic, the smaller the value of E indicates the higher the similarity of the segment and the candidate curve; select the candidate curve with the smallest total error E as the optimal typical load curve of the segment.

[0069] The typical matching value is an index reflecting the matching degree of the segment and the optimal typical curve, which is calculated by weighting the deviation loss value and the matching gain value:

[0070] Calculate the typical difference curve: ΔP i =P i actual-P i optimal, (i is the sampling point serial number in the segment, 1 to M, M is the number of sampling points), which constitutes the difference curve;

[0071] Calculate the deviation loss value D: adopt the segment linear function D=f(ΔPavg, ΔPmax), where ΔPavg=ΣΔP i / M is the average difference value of the difference curve, ΔPmax=max(|ΔP i|) maximum absolute difference, Pavg optimal is the optimal curve average value, the function rule is: if |ΔPavg|≤5%×Pavg optimal and ΔPmax≤10%×Pavg optimal, then D=0.9-10×|ΔPavg| / Pavg optimal-5×ΔPmax / Pavg optimal; if |ΔPavg|>5%×Pavg optimal and ΔPmax≤10%×Pavg optimal, then D=0.85-15×|ΔPavg| / Pavg optimal-5×ΔPmax / Pavg optimal; if |ΔPavg|≤5%×Pavg optimal and ΔPmax>10%×Pavg optimal, then D=0.85-10×|ΔPavg| / Pavg optimal-8×ΔPmax / Pavg optimal; if |ΔPavg|>5%×Pavg optimal and ΔPmax>10%×Pavg optimal, then D=0.8-15×|ΔPavg| / Pavg optimal-8×ΔPmax / Pavg optimal;

[0072] Calculate the matching gain value G: for each feature j, assign a gain contribution gj according to the value of εj (gj=wj×1.0 when εj≤5%; gj=wj×0.8 when 5%<εj≤10%; gj=wj×0.6 when 10%<εj≤15%; gj=wj×0.4 when εj>15%), calculate G by the following formula:

[0073]

[0074] wherein gj is the gain contribution of the jth feature; j

[0075] Calculate the typical matching value S according to G: adopt a weighted product model, and α and β are weights:

[0076] S=(G α )×(Dβ)

[0077] The rule recognition strategy further includes a model construction sub-strategy, and the model construction sub-strategy includes:

[0078] ​Step A41, identify typical load curve clusters of different electricity nodes at daily, weekly, and monthly time scales using an unsupervised learning algorithm based on density clustering, and generate corresponding multi-time labels; DBSCAN, a density-based unsupervised learning algorithm, can effectively handle non-spherical distributed data and noise points by identifying high-density regions in the data set as clusters. The parameters are set as follows: neighborhood radius ε = 0.1 × Pavg_history (for residential nodes), 0.08 × Pavg_history (for commercial nodes), and 0.05 × Pavg_history (for industrial auxiliary nodes), where Pavg_history is the average power consumption of the target node in the past 12 months; minimum sample size MinPts = 5. The clustering process is as follows: collect the load data of the target node in the past 12 months, and segment it according to the multi-time label; extract 5 matching features of each segment to form a feature vector; input the feature vector into the DBSCAN algorithm, and the algorithm automatically identifies clusters, each cluster is assigned a corresponding multi-time label.

[0079] Step A42, split the pre-set random test samples according to the typical load curve clusters based on the multi-time label to generate interpretation information; the random test samples are a segmented set randomly selected from the historical data at a proportion of 20%, which needs to cover all multi-time label combinations to ensure comprehensive testing. The splitting process is as follows: according to the multi-time label of the sample, it is assigned to the corresponding cluster of the label. The interpretation information is structured data describing the matching of the sample and the cluster, including the multi-time label combination of the sample, the cluster number to which it belongs, the Euclidean distance d between the sample feature vector and the cluster center feature vector, and the assignment result.

[0080] Step A43, evaluate whether the interpretation information meets the pre-set interpretation conditions, if it does, go to step A44; if not, generate cluster adjustment instructions based on the interpretation information to adjust the corresponding classical load curve cluster, and return to step A42; the interpretation conditions are the judgment criteria for model performance, specifically, the correct assignment rate R of the test samples under the same label combination is ≥ 90%, and the average Euclidean distance d between all sample feature vectors and the cluster center in the cluster is ≤ 0.5 × ε. The evaluation and adjustment process is as follows: calculate the correct assignment rate R of each label combination = the number of correctly assigned samples / the total number of samples in that label combination × 100%, and calculate the d of each cluster = Σd / the number of samples in the cluster; if R of all label combinations ≥ 90% and d of all clusters ≤ 0.5 × ε, the interpretation conditions are met, go to step A44; if not, generate cluster adjustment instructions: if R of a certain label combination < 90%, increase ε of the corresponding cluster of that label, increase Pavg_history or decrease MinPts each time; if d of a certain cluster > 0.5 × ε, split that cluster into two new clusters according to feature differences; after adjustment, return to step A42 to re-split the test samples and generate interpretation information until the interpretation conditions are met. avg avg avg avg ​​​​

[0081] Step A44, generating a corresponding typical load curve, matching feature and deviation impact mapping function according to each typical load curve cluster to generate the typical load feature model. The average power of all segments in the cluster at each sampling time is calculated, i.e. for K segments in the cluster, the power value of each sampling time i is P i,k typical i =∑P i The average value of the five matching features of all segments in the cluster is calculated as the matching feature value of the typical load curve; based on the difference data of all segments in the cluster and the typical load curve, the coefficients of the segment linear function are fitted by linear regression to ensure that the function can accurately quantify the impact of different deviation degrees on matching; the above three components are stored according to the multi-time label classification to form a complete typical load feature model, and the model is periodically retrained to adapt to the seasonal changes of load features.

[0082] The typical load feature model generates a typical difference curve according to the difference between the segment and the typical load curve, calculates the deviation reduction value through the deviation impact mapping function, extracts the corresponding matching feature in the segment according to the matching feature to obtain the corresponding matching gain value, and brings the deviation reduction value and the matching gain value into the preset matching evaluation algorithm to obtain the corresponding typical matching value.

[0083] Step A5, fitting each segment by a preset dynamic rule fitting sub-strategy to generate a corresponding rule fitting value;

[0084] The dynamic rule fitting sub-strategy includes:

[0085] Step A51, calling a reference load period that meets the conditions through a multi-time label; the reference load period refers to a historical load period that is consistent with the current segment label, has complete historical data and stable load, and the calling process is: filtering the historical periods with completely consistent labels from the historical database; calculating the historical periods, and selecting the candidate reference periods with Cv≤0.3; if the candidate periods are ≥3, selecting the three periods with the smallest deviation from the current segment Pavg; if 1-2, directly selecting; if 0, adjusting the month label to the adjacent season to reselect, until at least one reference period is obtained.

[0086] Step A52, establishing the period association between segments by the obtained reference load period, so that each segment has at least one associated segment to construct a corresponding period association group, and the period associations of segments in different period association groups are different; the period association refers to the load trend association between the current segment and the target segment: the historical reference period segment or the adjacent same label segment, which is determined by the Pearson correlation coefficient r, M is the number of sampling points, P i current, P i target are the i-th point power of the current and target segments respectively:

[0087]

[0088] wherein r ranges from -1 to 1, and when r≥0.8, it is determined that there is a periodic correlation. The periodic correlation group is constructed by taking the current segment as the core, and all the associated target segments are included in the same group. If the target segments are also associated, they are unified into a group, otherwise, they are independently set up into a group, ensuring that all segments in each group are associated with each other, and the current segment belongs to at least one group.

[0089] Step A53, the theoretical waveform of each segment is calculated by periodic correlation, and the periodic deviation waveform is obtained by calculating the deviation between the actual waveform and the theoretical waveform. The "theoretical waveform" refers to the weighted average standard waveform of all segments in the correlation group. In the calculation formula, K is the number of segments in the group, P k，i is the power of the ith point of the kth segment.

[0090]

[0091] The periodic deviation waveform refers to the power deviation between the current segment and the theoretical waveform. The deviation formula is as follows:

[0092] ΔP k，i = P i,当前 -P k，i

[0093] Step A54, the periodic deviation value of each periodic correlation group is calculated by a preset periodic law evaluation algorithm. The periodic deviation value Q quantifies the deviation degree of the segment from the periodic law. The calculation process of the periodic law evaluation algorithm is as follows: 1. Calculate the average absolute value of the deviation ΔPtheoreticalavg=Σ|ΔPtheoretical i | / M; 2. Calculate the standard deviation of the deviation:

[0094]

[0095] 3. Calculate the periodic deviation value:

[0096]

[0097] If Q>1, then Q=1.

[0098] Step A55, select the periodic correlation group with the smallest periodic deviation value, extract the corresponding periodic matching density, and determine the law fitting value according to the periodic matching density and the periodic deviation value. The periodic matching density ρ reflects the closeness of the correlation of the segments in the group. L is the logarithm of the segments in the group that satisfy r≥0.8, and C(K,2)=K×(K-1) / 2 is the total possible number of pairs:

[0099]

[0100] wherein the regularity fitting value T is calculated by the following formula, (1-Q) reflects the goodness of fit, and p reflects the group reliability: T = (1-Q) x p.

[0101] Step A6, generate a regularity matching vector with the classic matching value and the regularity fitting value as components, and obtain the corresponding electricity regularity image from the pre-set regularity image library according to the regularity matching vector. The regularity matching vector is a two-dimensional vector V = (S, T) composed of the typical matching value S and the regularity fitting value T, which fully reflects the matching of the segment and the typical characteristics, the periodic regularity. The regularity image library is a pre-set image set, and each image corresponds to the value range of V and the load characteristic description. The image matching process is: compare V with the value range in the image library, find the image containing V; if multiple images contain V, calculate the Euclidean distance between V and the center vector of each image, and select the image with the smallest distance; if V does not match any image, generate a special fluctuation type image, label the fluctuation characteristics and trigger key monitoring, and subsequently optimize the model through data accumulation to improve the image.

[0102] The power supply matching unit is configured to establish a power supply matching relationship based on the electricity regularity image, so that each power generation node is matched with at least one photovoltaic power generation unit, and each photovoltaic power generation unit is configured with a dynamic power supply strategy for each power generation node. The input data of the first unit needs to cover three types of core data, which are the electricity regularity image output by the electricity regularity analysis unit, the real-time output power and short-term output prediction data of the photovoltaic monitoring unit, and the real-time running state data uploaded by the energy storage unit. All input data need to be synchronized at a second level through the internal communication bus of the microgrid to ensure that the data timeliness meets the dynamic matching demand. The output content of the second unit needs to include executable power supply matching instructions, including photovoltaic power supply power distribution value of each power consumption node, target charge and discharge power of the energy storage unit, power supply adjustment coefficient of the adjustable load, and standby power supply trigger threshold of the important load. The output instruction format needs to be compatible with the communication protocol of the photovoltaic inverter energy storage controller load control module to ensure that the instruction can be directly parsed by the execution equipment. The third unit needs to have self-correction ability, that is, by collecting the matched data in real time, the parameters in the matching strategy are optimized in reverse to improve the long-term matching accuracy. At the same time, it needs to interact with the microgrid monitoring center to upload the matching results and the running state in real time, which is convenient for overall scheduling.

[0103] The power supply matching unit is configured with a power supply matching strategy for determining the power supply matching relationship, and the power supply matching strategy includes:

[0104] Step B1: Generate a power generation characteristic profile based on the historical power generation data and power generation characteristic parameters of the power generation unit. This step requires classifying the loads of all power-consuming nodes within the microgrid and sorting them according to matching priority scores. Loads with higher scores are given priority access to photovoltaic power. Matching priority score S priority The calculation needs to comprehensively consider the load importance weight W load Load stability index T, photovoltaic output stability index S pv_stab Three factors:

[0105] S priority =w1×W load +w2×T+w3×S pv_stab

[0106] In the formula, w1, w2, and w3 are the weights of the three factors, and w1 + w2 + w3 = 1, S pv_stab S is a photovoltaic power output stability indicator. pv_stab The fluctuation coefficient is calculated based on the photovoltaic power output fluctuation coefficient:

[0107] CV pv =σ pv / P pv_available , σ pv S represents the standard deviation of photovoltaic output. pv_stab =1-CV pv (CV pv ≤1). After the calculation is completed, all loads are set according to S. priority The loads are sorted from highest to lowest to form a "load matching priority list". Each load entry in the list includes a load number and a priority level (P). load_pred_corr S priority The value indicates that subsequent power allocation will be performed sequentially according to this list.

[0108] Step B2: Prioritize the electricity consumption pattern profile and power generation characteristic profile according to the preset classification criteria; calculate the current power supply and demand difference to determine whether there is a power shortage or a power surplus, providing a basis for subsequent power allocation. This is done using the following formula:

[0109] P supply_demand =P pv_available -P load_total_pred_corr

[0110] Among them, the power supply and demand difference ΔP supply_demand The calculation is based on the corrected photovoltaic available output and load power prediction values, ΔP supply_demand For the supply and demand difference, P pv_available To correct the available photovoltaic output, P load_total_pred_corr For all loads Pload_pred_corr The sum of .

[0111] When ΔP supply_demand When the power is greater than 0, it is determined to be a power surplus, and the surplus power ΔP surplus =ΔP supply_demand It needs to be stored through energy storage charging; when ΔP supply_demand When the value is less than 0, it is determined to be a power supply gap, and the gap power ΔP deficit =-ΔP supply_demand It needs to be supplemented through energy storage discharge or backup power; when |ΔP supply_demand |≤5%×P load_total_pred_corr When the supply and demand are in balance, there is no need to call up energy storage.

[0112] Step B3: Match the electricity consumption pattern profiles and power generation characteristic profiles of units with the same priority to assign them to the corresponding consumption matching groups. When a power supply surplus is determined, the goal of power allocation is to store the surplus photovoltaic output in energy storage units as much as possible, while avoiding exceeding the safe limit for energy storage SOC. This step is executed in two steps: the first step is to calculate the maximum rechargeable power P of the energy storage. ess_charge_max The second step is to determine the actual energy storage charging power P. ess_charge Maximum rechargeable energy storage power P ess_charge_max The current SOC value of the energy storage, the maximum charging power limit of the energy storage, and the charging efficiency need to be considered. The calculation is as follows:

[0113]

[0114] Among them, P ess_charge_max For maximum rechargeable power of energy storage, SOC upper C is the upper limit for safe charging of energy storage. ess P represents the rated energy storage capacity, Δt represents the matching cycle duration, and P represents the rated energy storage capacity. ess_charge_rated η is the rated charging power for energy storage. ess_charge This is the corrected energy storage charging efficiency.

[0115] Actual energy storage charging power P ess_charge The determination requires comparison of ΔP surplus With P ess_charge_max In the formula, when ΔP surplus ≤P ess_charge_max When, the energy storage fully absorbs the surplus power; when ΔP surplus >P ess_charge_max At that time, the energy storage is charged at its maximum rechargeable power, and the remaining surplus power ΔP curtailment =ΔP surplus -P ess_charge_max The output needs to be limited by the photovoltaic inverter, and the curtailment data should be recorded for subsequent strategy optimization.

[0116]

[0117] Step B4, obtaining the power consumption average of each power consumption node in the matching group and the power generation supply average of each power generation unit, and establishing the power supply matching relationship between the power consumption nodes and the power generation units according to the loss target constraint, so that each power consumption node is matched with at least one power generation unit and meets the preset power supply requirement constraint, the power supply requirement constraint being that the power supply demand difference between the power consumption nodes and the power generation units having the power supply matching relationship is within the preset reference supply-demand range, and the reference supply-demand range corresponding to each priority is different. When it is determined that there is a power supply gap, the power distribution needs to be performed in the order of "photovoltaic priority-energy storage supplement-load adjustment", to ensure that the power supply demand of important loads is met. The first step is photovoltaic power directional distribution, which allocates photovoltaic power to each load according to the "load matching priority list" from high to low, and the allocation rule is that for the priority i load, the allocated power P pv_allocate,i = min(P load_pred_corr,i , remaining photovoltaic power), the initial value of the remaining photovoltaic power being P pv_available , and after the distribution is completed, the remaining power supply gap ΔP deficit_1 = ΔP deficit - (P pv_available - remaining photovoltaic power) is calculated. The second step is energy storage discharge supplement, and the maximum dischargeable power P ess_discharge_max of the energy storage is calculated, P ess_discharge_max being the maximum dischargeable power of the energy storage, SOC lower being the lower limit of safe discharge of the energy storage, and η ess_discharge being the corrected energy storage discharge efficiency:

[0118]

[0119] The actual energy storage discharge power P ess_discharge = min(ΔP deficit_1 , P ess_discharge_max ), and the remaining power supply gap ΔP deficit_2 = ΔP deficit_1 - P ess_discharge after discharge. The third step is adjustable load adjustment, and for the adjustable load W load = 0.1 in the load matching priority list, the power supply power thereof is reduced according to the adjustment coefficient k adjust , and the calculation formula of the adjustment coefficient k adjust is ΔP deficit_2 = remaining power supply gap, and P adjustable_total is the sum of P load_pred_corr of all adjustable loads:

[0120]

[0121] When k adjustWhen <0.5, it means that the gap cannot be filled by adjusting the adjustable load only, and the standby power supply needs to be triggered to supply power, and the standby power supply power P backup = ΔP deficit_2 -(1-k adjust )×P adjustable_total , k adjust = 0.5. The power distribution result is converted into specific execution instructions, including the power distribution instructions of the photovoltaic inverter, the charge and discharge instructions of the energy storage controller, and the power adjustment instructions of the adjustable load. The power distribution instructions of the photovoltaic inverter need to be matched according to the priority list of the load, and the photovoltaic power P pv_allocate,i is distributed to each load, and the distribution value is sent to the load interface of the corresponding photovoltaic inverter.

[0122] The loss target constraint is configured with a loss target function to minimize the loss target value of the consumption matching group after the power supply matching relationship is established, and the loss target function calculates the loss target value based on a transmission loss value, a quality matching loss value, and an energy storage loss value. The transmission loss value reflects the loss in the transmission process, the quality matching loss value reflects the loss caused by the mismatch of power supply quality, and the energy storage loss value reflects the loss caused by the corresponding energy storage demand due to excessive power supply. The loss target constraint refers to quantifying the photovoltaic curtailment loss, energy storage charging and discharging loss, and power interruption loss in the microgrid operation to control the total loss below the preset threshold to ensure economic efficiency and reliability. First of all, it needs to be understood that there are multiple ways to combine the consumption matching group of the power supply matching relationship, that is, the purpose of minimizing the loss target value is to match the power supply quality and the power demand with higher matching degree and larger energy storage margin by weighting the transmission loss value, the quality matching loss value, and the energy storage loss value, so as to maximize the consumption of photovoltaic energy.

[0123] The dynamic power supply strategy includes:

[0124] Step C1, generating a corresponding power supply topology level according to the real-time acquired power data, environmental monitoring data and power load data by a preset topology evaluation algorithm, and configuring the power supply topology of the power consumption node according to the power supply topology level, the mapping relationship between the power supply topology level and the power supply topology is different under different priority; the power supply topology includes a first power supply topology, a second power supply topology, a third power supply topology, a fourth power supply topology and a fifth power supply topology, the first power supply topology is to connect only the photovoltaic unit of the matching group to the power supply loop of the power supply node, the second power supply topology includes connecting the photovoltaic unit of the matching group and the standby photovoltaic unit to the power supply loop of the power supply node, the third power supply topology includes connecting the photovoltaic unit of the matching group and the standby energy storage unit to the power supply loop of the power supply node, the fourth power supply topology includes connecting the photovoltaic unit of the matching group and the external power supply to the power supply loop, and the fifth power supply topology includes connecting the external power supply to the power supply loop. The "power consumption node priority" refers to the high priority node, which refers to the node such as hospital intensive care unit and industrial production line key equipment, the interruption of which will cause significant loss, the medium priority refers to the commercial node such as supermarket and office building, the interruption of which will affect the operation but the loss is controllable, and the low priority refers to the node such as residential lighting and ordinary household appliances, the interruption of which will have little influence. The preset topology evaluation algorithm needs to calculate the topology evaluation score based on three kinds of real-time data, and then divide the power supply topology level according to the score, and the specific steps are as follows:

[0125] First, collect real-time basic data: real-time photovoltaic output P pv , average photovoltaic output P in the last 10 minutes pv-avg10 , current light intensity G, average light intensity G in the last 1 hour avg60 , current power load P load , power load P in the last 5 minutes load-prev5 ; then calculate the key influence coefficient:

[0126] Photovoltaic real-time output fluctuation coefficient C pv , reflecting the stability of photovoltaic output, the formula is:

[0127]

[0128] Light intensity deviation rate C g , reflecting the potential influence of environment on photovoltaic output, the formula is:

[0129]

[0130] Power load mutation rate C load , reflecting the power load fluctuation, the formula is:

[0131]

[0132] The topology evaluation score S is calculated: according to the node priority allocation coefficient weight, the formula is:

[0133] High priority: S = 0.2 x C pv + 0.2 x C g + 0.6 x C load

[0134] Medium priority: S = 0.3 x C pv + 0.3 x C g + 0.4 x C load

[0135] Low priority: S = 0.4 x C pv + 0.4 x C g + 0.2 x C load .

[0136] Step C2, obtain the energy storage response constraint according to the power supply topology, determine the state data of the corresponding energy storage unit through the energy storage response constraint, and execute the corresponding execution sub-strategy according to the state data, and the energy storage response constraints of different dynamic power supply strategies are different. The execution sub-strategy includes executing the energy storage load mode, executing the energy storage mode, executing the load mode, and executing the energy storage cooperative supply mode. The energy storage load mode is that the photovoltaic unit supplies power to the energy storage unit and the power consumption node at the same time. The energy storage mode is that the photovoltaic unit only supplies power to the energy storage unit. The load mode is that the photovoltaic unit only supplies power to the power consumption node. The energy storage cooperative supply mode is that the energy storage unit and the photovoltaic unit supply power to the power consumption node at the same time. The core function is to adjust the charging and discharging of the energy storage to suppress the imbalance between supply and demand caused by photovoltaic output fluctuation and load change, and to prolong the service life of the energy storage. This mechanism needs to combine the SOC change trend of the energy storage, the photovoltaic output prediction curve, and the load power prediction curve to realize predictive charging and discharging, rather than passive response only after the imbalance between supply and demand. The energy storage response constraint refers to the energy storage operation parameter limit determined based on the current power supply topology, including the upper and lower limits of the SOC, the charging and discharging power limit, which needs to be calculated in combination with the topology characteristics, as follows:

[0137] Determine the energy storage response constraint parameter: let the rated charging and discharging power of the energy storage be P ess-rated , the current SOC be SOC current (%) and the different topology constraints be:

[0138] Pure consumption matching group photovoltaic:

[0139]

[0140] Photovoltaic with backup:

[0141]

[0142] Photovoltaic with backup energy storage:

[0143]

[0144] External grid-connected photovoltaic:

[0145]

[0146] Pure grid power supply:

[0147] SOC low = 35%

[0148] , SOC high = 80%, P ess-limit = P ess-rated × 0.5

[0149] (only when photovoltaic power generation cannot be performed);

[0150] Trigger the execution of the sub-strategy: collect the current charge and discharge power P ess-current of the energy storage, combine the constraints and real-time supply and demand trigger strategy:

[0151] Execute the energy storage load mode: when P pv > P load , SOC current < SOC high , and P ess-current < P ess-limit , the photovoltaic power is divided into two parts, the load power supplied is equal to P load , and the energy storage power supplied is P pv -P load ;

[0152] Execute the energy storage mode: when P pv > P load , SOC current ≤ SOC low + 10%, and P ess-current < P ess-limit , the photovoltaic power is entirely supplied to the energy storage, not exceeding P ess-limit , and the load is provided with basic power by the energy storage, not exceeding P load × 10%;

[0153] Execute the load mode: when |P pv -P load | ≤ 5% × P load , SOC current ∈ [SOC low , SOC high ], the photovoltaic power is entirely supplied to the load, and the energy storage is on standby without charging and discharging;

[0154] Execute the energy storage cooperative supply mode: when P pv < P load , SOCcurrent SOC low and |P ess-current |<P ess-limit , the energy storage discharging power is P load -P pv , and the photovoltaic supplies the load together.

[0155] Of course, the above is only a typical example of the present application, in addition to which the present application can have other various specific embodiments, and any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of the present application.

Claims

1. A photovoltaic power generation consumption system based on a distributed testing unit, deployed in a microgrid architecture, the microgrid architecture including power consumption nodes, photovoltaic power generation units, and external grid power, wherein the photovoltaic power generation units are equipped with photovoltaic inverter controllers, and the power consumption nodes are equipped with energy storage units, characterized in that, include: A distributed test unit cluster is deployed at each power consumption node to collect and upload power load data in real time. A photovoltaic monitoring unit cluster is deployed in the photovoltaic power generation unit to acquire output power data and environmental monitoring data of the photovoltaic power generation unit. A power distribution unit, wherein one side of the power distribution unit is coupled to a photovoltaic power generation unit and the other side is coupled to a power consumption node, and the power distribution unit is used to connect the power consumption node to the power supply circuit of the photovoltaic power generation unit according to the matching relationship between the photovoltaic power generation unit and the power consumption node. The electricity consumption pattern analysis unit is equipped with a pattern recognition strategy. The pattern recognition strategy is used to perform pattern recognition and feature extraction on the collected electricity load data at multiple time scales to generate an electricity consumption pattern profile. The power supply matching unit is used to establish a power supply matching relationship based on the power consumption pattern profile, so that each power generation node is matched with at least one photovoltaic power generation unit, and each photovoltaic power generation unit is configured with a dynamic power supply strategy for each power generation node.

2. The photovoltaic power generation consumption system based on a distributed testing unit according to claim 1, characterized in that, The pattern recognition strategy includes: Step A1: Receive the time series data of the electricity load of the target electricity node; Step A2: Preprocess the time series data, including imputation of missing values ​​and removal of outliers; Step A3: Segment the electricity load time series data using multiple time labels; Step A4: Match each segment with a preset typical load characteristic model to obtain the optimal typical load curve for each segment, and generate the typical matching value for each segment. Step A5: Fit each segment using a preset dynamic pattern fitting strategy and generate the corresponding pattern fitting value. Step A6: Generate a pattern matching vector using classic matching values ​​and pattern fitting values ​​as components, and obtain the corresponding electricity consumption pattern profile from the preset pattern profile library based on the pattern matching vector.

3. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 2, characterized in that, The pattern recognition strategy also includes a model building sub-strategy, which includes: Step A41: Using an unsupervised learning algorithm based on density clustering, identify typical load curve clusters of different electricity consumption nodes at daily, weekly, and monthly time scales, and generate corresponding multi-time labels; Step A42: Based on multiple time labels, split the preset random test samples according to the typical load curve cluster to generate explanatory information; Step A43: Evaluate whether the interpretation information meets the preset interpretation conditions. If it does, proceed to step A44; if it does not, generate a cluster adjustment instruction based on the interpretation information to adjust the corresponding classical load curve cluster, and return to step A42. Step A44: Generate the corresponding typical load curve, matching characteristics, and deviation influence mapping function for each typical load curve cluster to generate the typical load characteristic model.

4. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 3, characterized in that, The typical load characteristic model generates a typical difference curve by calculating the difference between the segmented electricity load time series data and the typical load curve. The deviation reduction value is obtained by calculating the deviation influence mapping function. The corresponding matching feature in the segment is extracted according to the matching feature to obtain the corresponding matching gain value. The deviation reduction value and the matching gain value are then used in a preset matching evaluation algorithm to obtain the corresponding typical matching value.

5. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 4, characterized in that, The dynamic law fitting sub-strategy includes: Step A51: Retrieve the baseline load cycle that meets the conditions through multiple time tags; Step A52: Establish periodic relationships between segments using the obtained baseline load cycle, so that each segment has at least one associated segment to construct a corresponding periodic relationship group. The periodic relationships of segments in different periodic relationship groups are different. Step A53: Calculate the theoretical waveform of each segment through periodic correlation and obtain the periodic deviation waveform by calculating the deviation from the actual waveform. Step A54: Calculate the periodic deviation value of each periodic group using a preset periodicity evaluation algorithm; Step A55: Select the periodic association group with the smallest periodic deviation value, extract the corresponding periodic matching density, where the periodic matching density is the number of periodic associations in the periodic association group divided by the number of segments, and determine the fitted value of the pattern based on the periodic matching density and the periodic deviation value.

6. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 1, characterized in that, The power supply matching unit is configured with a power supply matching strategy to determine the power supply matching relationship. The power supply matching strategy includes: Step B1: Generate a power generation characteristic profile based on the historical power generation data and power generation characteristic parameters of the power generation unit; Step B2: Prioritize the electricity consumption pattern profile and power generation characteristic profile according to the preset classification criteria; Step B3: Match the electricity consumption pattern profiles and power generation characteristic profiles of the same priority to classify them into the corresponding consumption matching groups; Step B4: Obtain the average power load of each power consumption node and the average power generation and supply of each power generation unit in the power consumption matching group, and establish the power supply matching relationship between the power consumption node and the power generation unit according to the loss target constraint, so that each power consumption node is matched with at least one power generation unit and meets the preset power supply requirement constraint. The power supply requirement constraint is that the power supply and demand difference between the power consumption node and the power generation unit with the power supply matching relationship is within the preset benchmark supply and demand range, and the benchmark supply and demand range corresponding to each priority is different.

7. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 6, characterized in that, The loss target constraint is configured with a loss target function to minimize the loss target value of the absorption matching group after the power supply matching relationship is established. The loss target function calculates the loss target value based on the transmission loss value, the quality matching loss value, and the energy storage loss value. The transmission loss value reflects the loss situation during the transmission process, the quality matching loss value reflects the loss situation caused by the power supply quality mismatch, and the energy storage loss value reflects the loss situation corresponding to the energy storage demand caused by the excessive power supply.

8. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 1, characterized in that, The dynamic power supply strategy includes: Step C1: Based on the real-time power data, environmental monitoring data and power load data, a preset topology evaluation algorithm is used to generate the corresponding power supply topology level, and the power supply topology of the power consumption node is configured according to the power supply topology level. Under different priorities, the mapping relationship between the power supply topology level and the power supply topology is different. Step C2: Obtain energy storage response constraints based on the power supply topology, determine the state data of the corresponding energy storage unit through the energy storage response constraints, and execute the corresponding execution sub-strategy based on the state data. The energy storage response constraints are different for different dynamic power supply strategies.

9. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 1, characterized in that, The power supply topology includes five types: Type I, Type II, Type III, Type IV, and Type V. Type I power supply topology is a power supply circuit that connects only the photovoltaic units of the absorption matching group to the power supply node. Type II power supply topology includes a power supply circuit that connects the photovoltaic units of the absorption matching group and a spare photovoltaic unit to the power supply node. Type III power supply topology includes a power supply circuit that connects the photovoltaic units of the absorption matching group and a spare energy storage unit to the power supply node. Type IV power supply topology includes a power supply circuit that connects the photovoltaic units of the absorption matching group and external mains power. Type V power supply topology includes a power supply circuit that connects external mains power.

10. A photovoltaic power generation consumption system based on a distributed testing unit according to claim 1, characterized in that, The execution sub-strategies include executing energy storage load mode, executing energy storage mode, executing load mode, and executing energy storage co-supply mode. The energy storage load mode is in which the photovoltaic unit supplies power to both the energy storage unit and the power consumption node. The energy storage mode is in which the photovoltaic unit supplies power to only the energy storage unit. The load mode is in which the photovoltaic unit supplies power to only the power consumption node. The energy storage co-supply mode is in which both the energy storage unit and the photovoltaic unit supply power to the power consumption node.

Images