A photovoltaic station area time sequence simulation method and system based on multi-modal data fusion

Abstract

The application discloses a kind of photovoltaic substation time sequence simulation method and system of multimodal data fusion, it is related to power system simulation and distribution substation intelligent modeling technical field.The method includes: obtaining substation multi-source heterogeneous data to unify space-time alignment, form multi-modal time sequence characteristic matrix;Based on night without photovoltaic measurement data inversion main branch equivalent topology;In daytime with light working condition, photovoltaic power flow is stripped from user end net power, constructs independent photovoltaic injection object and pure load object and is connected to equivalent topology node, forms photovoltaic-load collaborative simulation topology;With the topology and all-day multi-modal time sequence characteristic matrix as input, execute continuous time sequence state deduction, and according to the residual error of simulation voltage and measured voltage, online correction branch impedance parameter, output the all-day time sequence power flow deduction result of self-correction.The application can realize photovoltaic substation rapid modeling and time sequence power flow high-precision simulation under the condition that topology profile is incomplete, with strong engineering practicability.

Description

Technical Field

[0001] This invention relates to the field of power system simulation and intelligent modeling of distribution transformer areas, and in particular to a method and system for time-series simulation of photovoltaic transformer areas using multimodal data fusion. Background Technology

[0002] With the large-scale integration of distributed photovoltaic (PV) power into low-voltage distribution substations, the operating characteristics of these substations have undergone fundamental changes. The traditional operating mode, primarily based on unidirectional load supply, is gradually transforming into a bidirectional power flow operating mode driven by both PV output and user load. During peak PV output periods in the daytime, the voltage on the user side may rise significantly, and some lines may even experience power backflow; while at night, when there is no PV power, the substation reverts to pure load operation. This highly intermittent and volatile operating characteristic necessitates the establishment of time-series simulation models that accurately reflect the actual operating state of the substations for voltage limit analysis, power flow calculations, and operational assessments.

[0003] Currently, power flow simulation methods for transformer substations typically rely on complete and accurate low-voltage topology files, including detailed parameters such as user phase access information, line routing and length, branch connection relationships, and conductor types. However, in actual engineering projects, low-voltage transformer substations generally face problems such as a large number of users, complex wiring relationships, and frequent on-site modifications and user relocations. This often leads to inconsistencies or even serious omissions between the topology information recorded in marketing management systems and geographic information systems and the actual physical wiring. Under these conditions, traditional power flow calculation methods based on complete topologies lack a reliable structural foundation and struggle to accurately simulate the temporal variations in voltage distribution, line voltage drop, and power flow direction.

[0004] To alleviate the reliance on complete topology, recent studies have attempted to utilize smart meter measurement data for topology identification or impedance parameter estimation. However, most existing methods focus on identifying static topologies, failing to adequately consider the "contamination" caused by random fluctuations in photovoltaic output on the measurement signals. This makes it difficult to directly generalize the topology obtained under pure load conditions at night to daytime photovoltaic output scenarios. Furthermore, existing methods typically process measurement data and topology parameters separately, lacking a dynamic self-correction mechanism for time-series simulations, leading to significant accumulation of simulation errors over long periods.

[0005] In summary, there is an urgent need for a technical solution that can fully utilize multi-source measurement information to achieve rapid modeling and high-precision time-series simulation of photovoltaic power stations under conditions of incomplete topological archives. Summary of the Invention

[0006] To address this, this invention provides a method and system for time-series simulation of photovoltaic power distribution areas using multimodal data fusion. This system solves the problems in the prior art where incomplete topology files of low-voltage distribution areas, difficulty in effectively separating the influence of photovoltaic power output from measurements, and the lack of a dynamic self-correction mechanism lead to low accuracy in time-series power flow simulation and accumulation of errors over long periods.

[0007] To address the aforementioned technical problems, embodiments of the present invention provide a photovoltaic power station time-series simulation method based on multimodal data fusion, the method comprising: Acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area that includes static physical information, dynamic measurement information, and environmental information; Based on measurement data under no photovoltaic conditions at night, the equivalent topology of the transformer area is inverted using the voltage drop pattern. The equivalent topology includes at least the main line and the equivalent branches derived when the local power change and voltage fluctuation are mismatched. Under daytime conditions with sunlight, based on the environmental information in the multimodal time-series feature matrix of the transformer area, the photovoltaic power flow is separated from the net power at the user end to obtain independent photovoltaic injection objects and pure load objects, and the two are respectively connected to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology. Using the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as input, continuous time-series state deduction is performed to obtain the time-series simulation results of the voltage of each node and the power flow of each branch. Based on the residuals between the simulation results and the measured data, the equivalent branch impedance parameters are corrected online, and the self-corrected all-day time-series power flow deduction results are output.

[0008] Preferably, the method for acquiring multi-source heterogeneous data within a transformer area and performing unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area containing static physical information, dynamic measurement information, and environmental information includes: Acquire static physical data, dynamic measurement data, and environmental time-series data. The static physical data includes the relationship between the user and the transformer, the incomplete topology of the GIS, and the unit impedance parameters of the conductors. The dynamic measurement data includes the user's active power, reactive power, and voltage time-series data. The environmental time-series data includes irradiance, weather conditions, and ambient temperature. Using user ID and sampling time as a unified index, the static physical data, dynamic measurement data and environmental time series data are mapped to the same user-time coordinate to obtain the basic fusion features of each user at each time. Extract the no-photovoltaic period at night based on irradiance, calculate the reactive power-voltage sensitivity of each user during the no-photovoltaic period at night, and fuse the basic fusion feature with the reactive power-voltage sensitivity to obtain the multimodal fusion feature vector of each user at each time. The multimodal fusion feature vectors of all users at the same time are arranged according to the user dimension to obtain the multimodal time series feature matrix of the transformer area at that time, and then arranged in chronological order to form a sequence of multimodal time series feature matrices for the whole day.

[0009] Preferably, the method for retrieving the equivalent topology of a transformer area based on measurement data under nighttime photovoltaic-free conditions and utilizing voltage drop patterns includes: Extract voltage data for each user during the nighttime period without photovoltaic power generation, calculate the nighttime average voltage value for each user based on the rated voltage of the low-voltage side of the distribution transformer, and select the user with the lowest nighttime average voltage value as the farthest user. Based on the nighttime voltage drop, active power, reactive power, and conductor unit impedance parameters of the farthest user, the equivalent length of the main line is calculated in reverse, and a single main line is constructed with the distribution transformer as the starting point and the farthest user as the ending point. Based on the relative relationship between the nighttime average voltage values ​​of the remaining users and the nighttime average voltage values ​​of the farthest user, the remaining users are mapped to their corresponding electrical locations on the single trunk line. When the power variation and voltage fluctuation of users in a local area do not match on the single trunk line, an equivalent branch is derived at the corresponding electrical location, and users in the local area are assigned to the equivalent branch, generating an equivalent topology of the transformer area that includes the trunk line and at least one equivalent branch.

[0010] Preferably, when the power variation and voltage fluctuation of users in a local area do not match on the single trunk line, the method of deriving an equivalent branch at the corresponding electrical location includes: Calculate the correlation between the actual voltage timing of each user during the nighttime period without photovoltaic power and the estimated voltage timing at the mapped location on the main line, and include users with a correlation less than a set threshold into the branch derivation discrimination object. Local areas are divided according to the user's mapping location on the trunk line. The power throughput and average voltage fluctuation of each local area are calculated, and a power-voltage mismatch index is constructed. When the power-voltage mismatch index is greater than the set threshold, it is determined that there is a power throughput and voltage fluctuation mismatch in the local area, and an equivalent branch is derived at the average electrical location on the main line in the local area. The equivalent length of the derived equivalent branch is calculated based on the remaining voltage drop of each user in the local area, and the users in the local area are then separated from the trunk line and redistributed to the derived equivalent branch.

[0011] Preferably, the method for separating the photovoltaic power flow from the net power at the user end to obtain independent photovoltaic injection objects and pure load objects based on the environmental information in the multimodal time-series feature matrix of the transformer area under daytime lighting conditions includes: Extracting the daytime sunlight hours, and based on irradiance, ambient temperature, and photovoltaic installed capacity, calculating the theoretical photovoltaic active power output and photovoltaic reactive power output of each user during the daytime sunlight hours based on the photovoltaic inverter output mechanism model; Based on the net active power of each user during the daytime sunlight period and the theoretical photovoltaic active power output, the pure load active power demand after stripping the photovoltaic output is calculated, and the pure load reactive power demand is obtained by stripping the photovoltaic reactive power from the net reactive power. Using the topology nodes corresponding to each user as indexes, the extracted photovoltaic active power output and photovoltaic reactive power output are constructed into virtual photovoltaic injection objects, and the extracted pure load active power demand and pure load reactive power demand are constructed into pure load objects.

[0012] Preferably, the method of connecting the two to corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology includes: The virtual photovoltaic injection object and the pure load object are injected as dual-drive objects into the corresponding node ports of each user in the equivalent topology; Calculate the net injected active power and net injected reactive power of each user node based on the algebraic sum of the power of the virtual photovoltaic injection object and the pure load object at each user node. When the net injected active power is positive, it is determined that the user node injects active power into the distribution area; when the net injected active power is negative, it is determined that the user node absorbs active power from the distribution area. Using the equivalent topology as a spatial carrying skeleton, the virtual photovoltaic injection objects and pure load objects of all user nodes are attached to the corresponding nodes to form the photovoltaic-load collaborative simulation topology.

[0013] Preferably, the method for obtaining the time-series simulation results of each node voltage and each branch power flow by performing continuous time-series state deduction using the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as input includes: Based on the upstream and downstream relationships of each equivalent branch in the photovoltaic-load collaborative simulation topology, the active power flow and reactive power flow of each equivalent branch at each time point are calculated. Based on the impedance parameters of each equivalent branch and its active and reactive power flows, the voltage drop from the distribution transformer to each node is calculated, and the simulated voltage of each node at each moment is calculated with the rated voltage of the low-voltage side of the distribution transformer as the benchmark. Arrange the simulated node voltages at all times in chronological order to obtain the time-series simulation results of node voltages throughout the day. Arrange the active and reactive power flows of each branch at all times in chronological order to obtain the time-series simulation results of branch power flows throughout the day.

[0014] Preferably, the method for online correction of equivalent branch impedance parameters based on the residuals of simulation results and measured data, and outputting self-corrected all-day time-series power flow projection results, includes: The simulated voltage of each node at each time point is compared with the measured voltage of the same node at the same time point. The voltage residual of each node at each time point is calculated, and an evaluation index for the voltage simulation error throughout the day is constructed based on the voltage residual of all nodes and all time points. With the goal of minimizing the evaluation index of the all-day voltage simulation, the partial derivatives of the evaluation index with respect to the resistance and reactance of each equivalent branch are calculated, and the resistance and reactance of each equivalent branch are updated according to the set correction step size. The updated equivalent branch impedance parameters are re-substituted into the photovoltaic-load co-simulation topology, and the continuous time-series state deduction and impedance parameter update are repeated until the convergence criterion is met. The self-corrected all-day time-series power flow deduction results are then output.

[0015] Preferably, the method is executed independently by a local terminal device in the low-voltage distribution area. The local terminal device acquires dynamic measurement data collected by smart meters in the distribution area and environmental time-series data collected by locally deployed meteorological sensors, and completes the entire process calculation of topology inversion, photovoltaic stripping, time-series extrapolation and impedance self-correction locally, without relying on external servers or cloud platforms.

[0016] This invention also provides a photovoltaic power station time-series simulation system based on multimodal data fusion. This system is used to implement the aforementioned photovoltaic power station time-series simulation method based on multimodal data fusion, specifically including: The data acquisition and alignment module is used to acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area that includes static physical information, dynamic measurement information and environmental information. The topology inversion module is used to invert the equivalent topology of the transformer area based on measurement data under nighttime photovoltaic-free conditions and the voltage drop pattern. The equivalent topology includes at least the main line and the equivalent branches derived when the local power changes and voltage fluctuations are mismatched. The photovoltaic stripping and attaching module is used to strip the photovoltaic power flow from the net power of the user end under daytime sunlight conditions, based on the environmental information in the multimodal time-series feature matrix of the transformer area, to obtain independent photovoltaic injection objects and pure load objects, and attach the two to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology. The timing simulation and self-correction module is used to perform continuous timing state simulation with the photovoltaic-load co-simulation topology and the all-day multimodal timing feature matrix as input, to obtain the timing simulation results of the voltage of each node and the power flow of each branch, and to correct the equivalent branch impedance parameters online based on the residuals of the simulation results and the measured data, and output the self-corrected all-day timing power flow simulation results.

[0017] As can be seen from the above technical solutions, this invention application has the following beneficial effects: First, this invention breaks through the dependence of traditional power flow simulation on complete topology files. By utilizing measurement data and voltage drop patterns under nighttime conditions without photovoltaic power, it inverts the simplified topology of the transformer area, which includes the main line and equivalent branches. It can quickly establish a reliable simulation structure foundation even when the topology file is incomplete or missing. It is suitable for engineering scenarios where the actual low-voltage transformer area topology information is incomplete and users frequently change connections.

[0018] Second, this invention achieves refined decoupling and collaborative modeling of photovoltaics and loads. By introducing a photovoltaic inverter output mechanism model and combining irradiance and temperature characteristics, the photovoltaic power flow is separated from the net power measurement at the user end, and independent photovoltaic injection objects and pure load objects are constructed respectively. This effectively solves the problem of "pollution" of measurement signals caused by random fluctuations in photovoltaics, and enables the actual electricity consumption behavior of the load to be accurately characterized during daytime periods with sunlight.

[0019] Third, this invention establishes a closed-loop self-calibration mechanism of "simulation-measurement residual-impedance correction". By dynamically comparing the time-series simulation results with the measured voltage of the smart meter, the equivalent branch impedance parameters are corrected online. This significantly suppresses the cumulative effect of long-term simulation errors, ensuring that the time-series power flow projection results closely reflect the actual operating conditions of the transformer area. Actual verification shows that the average absolute percentage error can be reduced from 5.2% using traditional methods to 0.8%. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Referring to the drawings will make the features and advantages of the present invention clearer. The drawings are illustrative and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of a photovoltaic power station time-series simulation method based on multimodal data fusion provided in an embodiment of the present invention; Figure 2 This is a 24-hour node voltage timing curve diagram in an embodiment of the present invention; Figure 3This is a 24-hour branch power timing curve diagram in an embodiment of the present invention; Figure 4 This is a comparison chart of simulation errors between the method of the present invention and other existing methods in this embodiment of the invention; Figure 5 This is a block diagram of a photovoltaic substation time-series simulation system with multimodal data fusion provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1: To address the problems of low accuracy and long-term error accumulation in existing time-series power flow simulations caused by incomplete low-voltage distribution area topology files, difficulty in effectively separating the influence of photovoltaic output from measurements, and the lack of dynamic self-correction mechanisms, this invention proposes a multi-modal data fusion method for photovoltaic distribution area time-series simulation. Figure 1 As shown, the method includes the following steps: S1: Acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area containing static physical information, dynamic measurement information, and environmental information; S2: Based on the measurement data under the condition of no photovoltaic at night, the equivalent topology of the transformer area is inverted by using the voltage drop law. The equivalent topology includes at least the main line and the equivalent branch derived when the local power change and voltage fluctuation are mismatched. S3: Under daytime lighting conditions, based on the environmental information in the multimodal time-series feature matrix of the transformer area, the photovoltaic power flow is separated from the net power at the user end to obtain independent photovoltaic injection objects and pure load objects, and the two are respectively connected to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology. S4: Using the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as input, perform continuous time-series state deduction to obtain the time-series simulation results of the voltage of each node and the power flow of each branch. Based on the residuals between the simulation results and the measured data, correct the equivalent branch impedance parameters online and output the self-corrected all-day time-series power flow deduction results.

[0023] As can be seen from the above technical solution, this invention proposes a multimodal data fusion method for photovoltaic (PV) distribution area time-series simulation. This method first acquires multi-source heterogeneous data from the distribution area and unifies spatiotemporal alignment to form a multimodal time-series feature matrix, providing a unified and reliable data foundation for subsequent topology inversion, PV stripping, and time-series extrapolation. Based on this, the main trunk line is inverted using measurement data from nighttime PV-free conditions. The branch equivalent topology addresses the issue of incomplete topology files in low-voltage distribution areas and frequent user reconnections leading to a lack of reliable structural foundation for power flow calculations. Furthermore, under daytime sunny conditions, photovoltaic power flow is extracted from the net power at the user end based on environmental information, and independent photovoltaic injection objects and pure load objects are constructed and connected to the equivalent topology nodes to form a photovoltaic... The load co-simulation topology effectively solves the problem of "pollution" of measurement signals caused by random fluctuations in photovoltaics, resulting in daytime simulation distortion and difficulty in decoupling photovoltaics and loads. Finally, continuous time-series state deduction is performed using this co-simulation topology and the all-day multimodal time-series characteristic matrix as inputs. The equivalent branch impedance parameters are corrected online based on the residual between the simulated voltage and the measured voltage, realizing closed-loop self-correction of simulation errors over long periods and outputting the all-day time-series power flow deduction results.

[0024] Furthermore, in step S1, multi-source heterogeneous data within the transformer area are acquired and uniformly spatiotemporally aligned to form a transformer area multimodal temporal feature matrix containing static physical information, dynamic measurement information, and environmental information.

[0025] Specifically, firstly, multi-source heterogeneous data of the target photovoltaic area is acquired, including static physical data, dynamic measurement data, and environmental time-series data. The static physical data acquired includes the user-transformer relationship table from the marketing system, incomplete original GIS topology remnants, and unit impedance parameters of conductors with different diameters. The user-transformer relationship table is used to determine the affiliation between users and distribution transformers; the original GIS topology remnants provide information on existing line connections; and the conductor unit impedance parameters characterize the resistance and reactance characteristics of lines with different diameters.

[0026] Dynamic measurement data includes active power, reactive power, and voltage time-series data collected by smart meters on the user side. Assume there are a total of [number missing] [units missing] within the transformer area. The user, the first individual users at any time The active power, reactive power, and voltage are expressed as follows: , , .

[0027] Environmental time-series data includes micrometeorological data within the target area's coverage, primarily including irradiance, weather conditions, and ambient temperature. (Set time...) Irradiance, weather conditions, and ambient temperature are expressed as follows: , , .

[0028] The marketing system's user-transformer relationship table, incomplete GIS original topology fragments, and conductor unit impedance parameters are structured to form user-level static physical characteristics. For the first... For each user, construct a static physical feature vector: , in, Represents the static physical feature vector. Indicates the first User ownership characteristics Indicates the first Location association features of individual users in GIS residual maps Indicates the first The unit resistance of the user-associated line. Indicates the first The unit reactance of the user-associated line.

[0029] Through the above processing, the originally unstructured household-transformer relationships, incomplete topological information, and conductor parameters are transformed into calculable static physical characteristics.

[0030] Then, the active power, reactive power, and voltage data collected by the smart meters on the user side are processed into time series data to form user measurement characteristics. individual users at any time The dynamic measurement feature vector is represented as: , in, For the first individual users at any time The dynamic measurement feature vector.

[0031] By arranging the measurement data of all users at the same time, a transformer area measurement matrix is ​​obtained. This matrix is ​​used to describe the target transformer area at time [time value missing]. Full user running status: , in, For a moment The measurement matrix of the station area.

[0032] The micro-meteorological data within the target area's coverage area are processed over time to form an environmental feature vector. (Time) The environmental feature vector is represented as follows: , in, For a moment The environmental feature vector.

[0033] This environmental feature is used to reflect the external influencing factors of photovoltaic power output changes, providing environmental constraints for subsequent photovoltaic power station time-series simulations.

[0034] Next, we align the time and space dimensions. Using the smart meter sampling time as a benchmark, we unify static physical data, dynamic measurement data, and environmental time-series data onto the same time coordinate. For dynamic measurement data, we directly use the smart meter sampling time. Arrange the data. For environmental data, irradiance, weather conditions, and ambient temperature are arranged according to the same sampling time. Matching is performed. For static physical data, since the household-transformation relationships, GIS residual maps, and traverse parameters remain unchanged throughout the simulation period, they are copied to each sampling time. After time-dimensional alignment, the... individual users at any time It simultaneously possesses static physical characteristics, dynamic measurement characteristics, and environmental characteristics.

[0035] Using the user ID as a unified spatial index, the system matches user IDs, smart meter IDs, GIS residual map node IDs, and meteorological coverage areas within the marketing system. For the first... For each user, their household-transformer relationships, GIS residual map locations, conductor parameters, meter measurement data, and meteorological data are uniformly mapped to the same user index. After spatial alignment, the first... The data for each user includes: , , , , , , , , , This enables a consistent representation of static physical data, dynamic measurement data, and environmental data in the user space dimension.

[0036] The periods of nighttime without photovoltaic interference are extracted based on irradiance data. Let the set of nighttime periods without photovoltaic interference be: , in, This indicates the period of nighttime when there is no photovoltaic power. Indicates time The photovoltaic irradiance is zero. During this period, the user voltage is mainly affected by the load power and line impedance, and is not affected by the photovoltaic reverse output. Therefore, it can be used to extract initial characteristics reflecting the user's spatial location.

[0037] During nighttime periods without photovoltaic power, calculate the sensitivity between reactive power changes and voltage changes for each user. For the first... For each user, calculate their nighttime reactive power-voltage sensitivity: , in, Indicates the first Reactive power-voltage sensitivity of individual users Indicates the first during the nighttime period Voltage change for each user Indicates the first during the nighttime period Changes in reactive power for each user.

[0038] Specifically, and It can be represented as: , , in, and These are two sampling times during the nighttime period when there is no photovoltaic power. The closer the user's location is to the end of the line, the more pronounced the voltage response to reactive power changes. Therefore, It can be used as an initial spatial embedding feature for users under unknown topological conditions.

[0039] During the nighttime periods without photovoltaic power, extract the nighttime voltage statistics for each user, including the minimum nighttime voltage, the average nighttime voltage, and the nighttime voltage fluctuation. The lowest nighttime voltage for each user is expressed as follows: ; No. The average nighttime voltage for each user is expressed as follows: ; No. The nighttime voltage fluctuation of an individual user is expressed as follows: ; in, Indicates the first The lowest overnight voltage for each user Indicates the first Average nighttime voltage for each user Indicates the first The nighttime voltage fluctuation of each user. The above characteristics are used to help determine the relative distance between the user and the distribution transformer.

[0040] By fusing static physical characteristics, dynamic measurement characteristics, environmental characteristics, nighttime reactive power-voltage sensitivity characteristics, and nighttime voltage statistical characteristics, a third... individual users at any time Multimodal fusion feature vector: , in, Indicates the first individual users at any time The multimodal fusion feature vector contains the user's static physical attributes, real-time running status, external environmental status, and nighttime spatial location-sensitive features.

[0041] Finally, the multimodal fusion feature vectors of all users at the same time are arranged to obtain the time-varying feature vectors. Multimodal time-series feature matrix of the transformer area: , in, Indicates time The multimodal time-series feature matrix of the transformer area is obtained. Further, arranged in chronological order, the multimodal time-series feature sequence of the transformer area within the complete simulation cycle is obtained: , in, This represents the results of multimodal temporal features of the transformer area that are aligned in both the time and spatial dimensions.

[0042] Furthermore, in step S2, based on the measurement data under the condition of no photovoltaic power at night, the equivalent topology of the transformer area is inverted using the voltage drop law. The equivalent topology includes at least the main line and the equivalent branches derived when the local power change and voltage fluctuation are mismatched.

[0043] Specifically, based on the multimodal temporal feature matrix Extracting nighttime periods without light The system collects time-series data on voltage, active power, and reactive power for all users and reads the unit impedance parameters of the conductors. During the nighttime period when there is no sunlight, the photovoltaic output is zero, and the voltage in the distribution area is mainly affected by user load and line impedance. Therefore, this period is used to invert the electrical location of users and the equivalent topology.

[0044] The rated voltage of the low-voltage side of the distribution transformer Based on this, the nighttime voltage for each user is calibrated. individual users at any time The voltage rating is: , in, Indicates the first individual users at any time The voltage value. Calculate the first... Average voltage rating for each user during nighttime (darkness) periods: , in, Indicates the first Average nighttime voltage rating for each user This indicates the number of sampling points during the dark period at night.

[0045] Based on the voltage drop pattern under pure load conditions at night, the user with the lowest average nighttime voltage value is selected as the furthest user in the transformer area. Furthest User Number satisfy: , in, This represents a collection of users in the same area. This indicates the number of the furthest user. Since there is no reverse voltage rise from photovoltaic power during the nighttime period when there is no sunlight, the lower the user's voltage value, the more significantly the user is affected by the line impedance voltage drop. Therefore, the user with the lowest voltage value is defined as the critical node at the end of the main line.

[0046] Based on the low-voltage side voltage of the distribution transformer, the nighttime voltage of the furthest user, active power, reactive power, and conductor unit impedance parameters, the equivalent line length between the distribution transformer and the furthest user is calculated in reverse. Assume the unit resistance of the main line is... The unit reactance is The most remote user At any moment The voltage drop is: , in, For the most remote user At any moment The voltage drop.

[0047] Based on the approximate relationship of voltage drop in low-voltage lines: , Thus, the time is obtained. Equivalent length of the main trunk line below: , The equivalent physical length of the main line is obtained by averaging all sampling points during the nighttime darkness period: , in, It represents the equivalent physical length of the main line between the distribution transformer and the farthest user.

[0048] Taking the low-voltage side node of the distribution transformer as the starting point of the main line, and the furthest user as the starting point... As the terminus of the main trunk line, a construction length of [length missing] is [length missing]. A single main trunk line model. The location of the distribution transformer and the location of the farthest user are defined as: , , in, Indicates the starting point of the main route. This indicates the location of the furthest user on the main line.

[0049] Based on the relative relationship between the average nighttime voltage values ​​of the remaining users and the average nighttime voltage value of the furthest user, all users are initially mapped onto the trunk line. The initial electrical position of each user on the trunk line is as follows: , in, Indicates the first The initial mapping position of each user on the backbone line. Indicates the first Average nighttime voltage rating for each user This represents the average nighttime voltage value for the furthest user. The closer the user's average nighttime voltage value is to 1, the closer the user is to the distribution transformer; the closer the user's average nighttime voltage value is to... At that time, the closer the user is to the end of the main line.

[0050] Based on the user's initial mapping location on the trunk line, estimate the nighttime voltage at the corresponding location on the trunk line. The estimated trunk voltage at each user-mapped location is: , in, Indicates the first User mapping location Estimated voltage of the main line at the location.

[0051] Calculate the first Correlation between the actual nighttime voltage timing of an individual user and the voltage timing of its mapped location on the main line: , in, Indicates the first Voltage timing correlation between each user and the corresponding location on the main line Indicates the first Average nighttime voltage for each user This represents the average nighttime voltage at the location mapped from the main line. Greater than the set threshold At that time, retain the user At the main line location The connection relationship on; when Less than At that time, the user Included in the branch road derivation discrimination objects.

[0052] Based on the user's initial mapping location on the backbone, the backbone is divided into several local regions. Let the first region be... The user set in a local area is Calculate the first... A local region at time Power throughput: , in, For the first A local region at time Power throughput.

[0053] Calculate the first A local region at time Average voltage fluctuation: , in, For the first A local region at time The average voltage fluctuation.

[0054] Constructing a power throughput and voltage fluctuation asymmetry metric: , in, Indicates the first The degree of power-voltage asymmetry in a local area This represents the scaling factor between power throughput and voltage fluctuation. To set a threshold, when the condition is met At that time, the judgment of the first In some local areas, there is an asymmetry between power throughput and voltage fluctuation, which requires activation of the branch derived operator.

[0055] When the When a local area meets the conditions for branch line derivation, the average electrical position of that area on the main line is used as the branch line connection point: , in, Indicates the first The connection points of each branch on the main line. The residual voltage drop of the user within the branch relative to the connection point is: , in, Indicates the location of the main line At the moment The voltage estimate. The first voltage drop is calculated by back-calculating the remaining voltage drop. Equivalent length of the branch: , in, Indicates the first The equivalent length of a derived branch.

[0056] For the set of users that trigger branch derivation It was separated from its original main trunk position and redistributed to the first... On a side road. The... The user in the first The location of the branch road is: , in, Indicates the first The user in the first Equivalent position on the branch road Indicates the first The average nighttime voltage at the junction of the branch circuits, Indicates the first The lowest average nighttime voltage for users within a branch. The greater the voltage drop of a user, the closer its location is to the end of the branch; the smaller the voltage drop of a user, the closer its location is to the branch junction.

[0057] Power throughput and voltage fluctuation asymmetry are sequentially assessed for all local areas along the main line, and branch derivation operations are performed at locations that meet the conditions to obtain the equivalent topology model of the transformer area "main line-multiple branches" (i.e., the graph theory representation of the aforementioned equivalent topology): , in, This is a topological model of the equivalent structure of the "main trunk-multiple branch roads" in the transformer substation area. This represents the set of nodes in the equivalent topology, including distribution transformer nodes, trunk line nodes, branch connection nodes, and user nodes. This represents the set of edges in the equivalent topology, including the main path edges and derived branch edges.

[0058] Furthermore, in step S3, under daytime light conditions, based on the environmental information in the multimodal time-series feature matrix of the transformer area, the photovoltaic power flow is separated from the net power at the user end to obtain independent photovoltaic injection objects and pure load objects, and the two are respectively connected to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology.

[0059] Specifically, the daytime hours with sunlight are extracted as the calculation window for photovoltaic power output stripping. Let the set of daytime hours with sunlight be: , in, This indicates a gathering during the daytime when there is light. Indicates time Irradiance. At time Next, the The net active power and net reactive power collected by each user's smart meter are respectively expressed as: , ,in Indicates the first Net active power of each user terminal Indicates the first Net reactive power of each user terminal.

[0060] A physical output mechanism model for photovoltaic inverters is introduced, and the theoretical photovoltaic active power output at the nodes is calculated based on irradiance, ambient temperature, and temperature coefficient. individual users at any time The operating temperature of a photovoltaic module is expressed as: , in, Indicates the first individual users at any time The operating temperature of photovoltaic modules Indicates time Ambient temperature, This represents the coefficient of influence of irradiance on the temperature rise of the component.

[0061] No. individual users at any time The active power output of photovoltaic theory is expressed as: , in, Indicates the first individual users at any time The photovoltaic theory has contributed significantly. Indicates the first Photovoltaic installed capacity per user This represents the standard test irradiance (usually taken as 1000 W / m²). This represents the temperature power coefficient of a photovoltaic module (usually taken as -0.0035 / ℃). This indicates the standard test temperature (usually 25℃).

[0062] After considering the inverter conversion efficiency, the first individual users at any time The active power output of photovoltaic power is: , in, Indicates the first Conversion efficiency of individual user photovoltaic inverters.

[0063] During the daytime hours when sunlight is available, the net active power measured by the user's smart meter is jointly determined by the load's active power demand and the photovoltaic active power output, and their relationship is expressed as follows: , in, Indicates the first individual users at any time The pure load active power demand, Indicates the first individual users at any time The photovoltaic system is productive. When When it increases, The decrease manifests as a negative power flow caused by photovoltaic power generation. Therefore, the first... individual users at any time The negative photovoltaic power flow is: , in, This indicates the negative active power flow caused by photovoltaic power generation.

[0064] Based on the daytime net active power and the physical output model of photovoltaic inverters, the photovoltaic output component in the net power at the user end is separated to obtain the pure load active power demand. individual users at any time The active power demand of the pure load is: , in, This indicates the active power demand of the pure load after the photovoltaic output is removed.

[0065] For reactive power, the photovoltaic reactive power injection relationship is established according to the inverter power factor constraint. Let the first... The power factor of the photovoltaic inverter for each user is Then the reactive power output of photovoltaic power generation is expressed as: , in, Indicates the first individual users at any time The reactive power output of photovoltaic power Indicates the first The power factor angle of a user's photovoltaic inverter.

[0066] The user's pure load reactive power demand is expressed as: , in, This indicates the reactive power demand of the pure load after removing the reactive power output of photovoltaic power.

[0067] Based on the photovoltaic active power output and photovoltaic reactive power output obtained from the stripping, in the first... A virtual photovoltaic injection object is constructed at the node corresponding to each user. The virtual photovoltaic injection object for an individual user is represented as follows: , in, Indicates the first Virtual photovoltaic injection objects at each user node Indicates the first This object represents the corresponding node of a user in the topology graph model. It is used to characterize the dynamic injection behavior of photovoltaic power at user nodes under daytime sunlight conditions.

[0068] Based on the stripped-off pure load active power demand and pure load reactive power demand, construct the first... Pure load demand objects at each user node. The pure load requirement object for a single user is represented as follows: , in, Indicates the first The pure load demand object at each user node. This object independently expresses the user's actual electricity demand from the net power measurement results.

[0069] The virtual photovoltaic injection object and the pure load demand object are treated as dual-drive objects and injected into the corresponding user node ports of the topology graph model. For the first... For each user node, the net active and reactive power injected into the node are expressed as follows: , , in, Indicates the first Each user node at time Net active power injection, Indicates the first Each user node at time The net reactive power injection. When When, the node injects active power into the transformer area; when At that time, the node absorbs active power from the transformer area.

[0070] Using the nighttime pure load topology as the spatial load-bearing skeleton, the first Each user's virtual photovoltaic injection object and pure load demand object are attached to the node. .node The daytime object attachment relationship is represented as follows: , in, Indicates the first The corresponding nodes of each user in the topology graph. Through this process, the structural topology derived under the nighttime pure load mode is extended to the photovoltaic-load co-simulation topology under daytime light conditions.

[0071] After performing photovoltaic object stripping, load object construction, and dual-drive injection on all user nodes, a spatiotemporal collaborative simulation topology for the transformer substations with independent dynamic photovoltaic objects is formed. This topology is represented as follows: , , , in, For the spatiotemporal collaborative simulation topology of the transformer area, This represents the set of virtual photovoltaic injection objects for all user nodes. This represents the set of pure load requirements objects for all user nodes.

[0072] Furthermore, in step S4, the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix are used as inputs to perform continuous time-series state deduction, obtain the time-series simulation results of the voltage of each node and the power flow of each branch, and correct the equivalent branch impedance parameters online based on the residuals of the simulation results and the measured data, and output the self-corrected all-day time-series power flow deduction results.

[0073] Specifically, the 24-hour time-series input is represented as follows: , in, Represents the multimodal time-series feature matrix throughout the day. Indicates time The multimodal characteristics of the transformer area, This indicates the total number of sampling points throughout the day.

[0074] Establish equivalent impedance parameters for each branch in the aforementioned output topology. The equivalent impedance of the e-th branch is expressed as: , in, Indicates the first The equivalent impedance of the branch is Indicates the first The equivalent resistance of the branch, Indicates the first The equivalent reactance of the branch circuit, It represents the imaginary unit.

[0075] Abandoning the traditional single-point power flow calculation method, the entire day's operation process is represented as a continuous temporal state evolution process. Each node at time... The voltage change is expressed as: , in, Indicates the first Each node at time... The amount of voltage change. Indicates the distance from the distribution transformer to the node. The equivalent set of branches, Indicates the first a side road at all times active power, Indicates the first a side road at all times The reactive power.

[0076] No. Each node at time... The simulated voltage is: , in, Indicates the first Each node at time... The simulated voltage.

[0077] Based on the upstream and downstream relationships of the equivalent topology, calculate the active and reactive power flow of each branch at each time step. a side road at all times The active and reactive power flows are represented as follows: , , in, Indicates the first The set of nodes downstream of each branch. Through the above calculations, the power flow time-series simulation results for each branch throughout the day can be obtained: , , in, This represents the simulation results of the active power flow timing of the branch throughout the day. This represents the simulation results of the reactive power flow timing of the branch throughout the day.

[0078] The simulated output voltage is compared with the voltage measured by a real smart meter, and the dynamic residual is calculated. Each node at time... The voltage residual is expressed as: , in, Indicates the first Each node at time... Voltage residual, Indicates the first The actual measured voltage of each node.

[0079] The evaluation index for the voltage residual throughout the day is expressed as follows: , in, This represents the voltage simulation error throughout the day.

[0080] Based on the voltage dynamic residual, the impedance parameters of the equivalent branch are adaptively fine-tuned. The resistance update formula for the e-th branch is: ; No. The reactance update formula for this branch is: , in, Indicates the number of self-correction iterations. Indicates the resistance correction step size. This indicates the reactance correction step size.

[0081] The updated branch impedance is: .

[0082] The convergence of the self-correction process is determined by the error change between two consecutive iterations. The convergence criterion is: , in, Indicates the first Error of the next iteration Indicates the first Error of the next iteration This represents the convergence threshold. When the convergence criterion is met, parameter updates are stopped; when the convergence criterion is not met, the updated impedance parameters are substituted back into the simulation topology, and the timing state evolution calculation continues.

[0083] After completing closed-loop self-calibration, the output shows the 24-hour node voltage and power timing sequence projections of the actual operating state of the mapping area. The projection results are expressed as follows: , in, For the results of the deduction, This represents the simulation results of node voltage timing throughout the day.

[0084] The method proposed in this invention is executed independently by a local terminal device in a low-voltage distribution area. The local terminal device acquires dynamic measurement data collected by smart meters in the distribution area and environmental time-series data collected by locally deployed meteorological sensors, and completes the entire process calculation of topology inversion, photovoltaic stripping, time-series extrapolation and impedance self-correction locally, without relying on external servers or cloud platforms.

[0085] This method is applied to the actual power flow simulation of low-voltage photovoltaic substations. For example... Figure 2 As shown, the simulated voltage curve of this invention has a high degree of consistency with the actual measured voltage curve, and can accurately reflect the voltage fluctuation characteristics caused by changes in photovoltaic output and load. The node voltage time-series curve accurately presents the nighttime load voltage drop, the voltage rise caused by midday photovoltaic output, and the voltage drop during the evening peak; as shown... Figure 3 As shown, this invention can effectively characterize the daily net load evolution pattern of a photovoltaic (PV) distribution area, with a significant load trough occurring during midday due to PV power generation, and the load reaching its maximum during the evening peak period. Figure 4 As shown, compared with traditional single-line models and modeling methods based on incomplete GIS information, the equivalent topology inversion method proposed in this invention significantly reduces the time-series simulation error, with the average absolute percentage error decreasing from 5.2% to 0.8%, verifying the modeling accuracy and engineering practicality of the proposed method under the condition of lacking complete topology archives.

[0086] Example 2: This invention provides a photovoltaic (PV) power grid time-series simulation system based on multimodal data fusion. This system is used to implement the multimodal data fusion method for PV power grid time-series simulation described in Embodiment 1 above. Figure 5 As shown, it specifically includes: The data acquisition and alignment module is used to acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area that includes static physical information, dynamic measurement information, and environmental information. This module is specifically used for: acquiring static physical data, dynamic measurement data, and environmental time-series data. The static physical data includes user-transformer relationships, incomplete GIS topology, and conductor unit impedance parameters. The dynamic measurement data includes user active power, reactive power, and voltage time-series data. The environmental time-series data includes irradiance, weather conditions, and ambient temperature. Using user ID and sampling time as a unified index, the static physical data, dynamic measurement data, and environmental time-series data are mapped to the same user-time coordinate to obtain the basic fusion features of each user at each time. Based on irradiance, the module extracts the no-PV period at night, calculates the reactive power-voltage sensitivity of each user during the no-PV period at night, and fuses the basic fusion features with the reactive power-voltage sensitivity to obtain the multimodal fusion feature vector of each user at each time. The module arranges the multimodal fusion feature vectors of all users at the same time according to the user dimension to obtain the multimodal time-series feature matrix of the transformer area at that time, and arranges them in chronological order to form a sequence of multimodal time-series feature matrices for the entire day.

[0087] The topology inversion module is used to invert the equivalent topology of the transformer area based on measurement data under nighttime conditions without photovoltaic power, using the voltage drop pattern. The equivalent topology includes at least the main line and the equivalent branches derived when there is a mismatch between local power changes and voltage fluctuations. This module is specifically used for: extracting voltage data of each user during the nighttime period without photovoltaic power; calculating the nighttime average voltage value of each user based on the rated voltage of the low-voltage side of the distribution transformer; selecting the user with the lowest nighttime average voltage value as the farthest user; calculating the equivalent length of the main line based on the nighttime voltage drop, active power, reactive power, and conductor unit impedance parameters of the farthest user; constructing a single main line with the distribution transformer as the starting point and the farthest user as the ending point; mapping the remaining users to their corresponding electrical locations on the single main line based on the relative relationship between the nighttime average voltage values ​​of the other users and the nighttime average voltage value of the farthest user; when the power changes and voltage fluctuations of users in a local area do not match on the single main line, deriving equivalent branches at the corresponding electrical locations, allocating users in the local area to the equivalent branches, and generating an equivalent topology structure of the transformer substation that includes the main line and at least one equivalent branch.

[0088] The photovoltaic (PV) stripping and connection module is used to strip the PV power flow from the net power at the user end under daytime sunny conditions, based on environmental information in the multimodal time-series characteristic matrix of the transformer area, to obtain independent PV injection objects and pure load objects. These are then connected to the corresponding nodes in the equivalent topology, forming a PV-load collaborative simulation topology. Specifically, this module is used to: extract the daytime sunny periods; calculate the theoretical PV active power output and PV reactive power output for each user during the daytime sunny periods based on irradiance, ambient temperature, and PV installed capacity, using a PV inverter output mechanism model; calculate the pure load active power demand after stripping PV power output based on the net active power at the user end and the theoretical PV active power output during the daytime sunny periods, and obtain the pure load reactive power demand from the net reactive power at the user end based on the PV reactive power output; and construct a virtual PV system using the topology nodes corresponding to each user as indices, combining the stripped PV active power output and PV reactive power output. The injection objects are constructed by extracting the pure load active power demand and pure load reactive power demand into pure load objects; the virtual photovoltaic injection objects and the pure load objects are injected as dual-drive objects into the corresponding node ports of each user in the equivalent topology; the net injected active power and net injected reactive power of each user node are calculated based on the algebraic sum of the power of the virtual photovoltaic injection objects and pure load objects at each user node; the equivalent topology is used as the spatial carrying skeleton, and the virtual photovoltaic injection objects and pure load objects of all user nodes are attached to the corresponding nodes to form the photovoltaic-load collaborative simulation topology.

[0089] The time-series simulation and self-correction module is used to perform continuous time-series state simulations based on the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as inputs. It obtains the time-series simulation results of the voltage at each node and the power flow of each branch, and corrects the equivalent branch impedance parameters online based on the residuals between the simulation results and measured data, outputting the self-corrected all-day time-series power flow simulation results. Specifically, this module is used to: calculate the active and reactive power flows of each equivalent branch at each time step based on the upstream and downstream relationships of each equivalent branch in the photovoltaic-load co-simulation topology; calculate the voltage drop from the distribution transformer to each node based on the impedance parameters of each equivalent branch and its active and reactive power flows, and calculate the simulated voltage of each node at each time step using the rated voltage of the low-voltage side of the distribution transformer as a reference; arrange the node simulated voltages at all times in chronological order to obtain the all-day node voltage time-series simulation results; arrange the active and reactive power flows of each branch at all times in chronological order to obtain the all-day branch power flow time-series simulation results; and calculate the voltage drop from each node at all times... The simulated voltage is compared with the measured voltage at the same node at the same time, and the voltage residual of each node at each time is calculated. Based on the voltage residuals of all nodes and all times, a full-day voltage simulation error evaluation index is constructed. With the goal of minimizing the full-day voltage simulation error evaluation index, the partial derivatives of the evaluation index with respect to the resistance and reactance of each equivalent branch are calculated respectively. The resistance and reactance of each equivalent branch are updated according to the set correction step size. The updated equivalent branch impedance parameters are substituted back into the photovoltaic-load co-simulation topology, and the continuous time-series state deduction and impedance parameter update are repeated until the convergence criterion is met. The self-corrected full-day time-series power flow deduction result is output.

[0090] This embodiment provides a photovoltaic (PV) power grid time-series simulation system based on multimodal data fusion, used to implement the aforementioned multimodal data fusion PV power grid time-series simulation method. Therefore, the specific implementation of the multimodal data fusion PV power grid time-series simulation system can be found in the embodiment section of the multimodal data fusion PV power grid time-series simulation method above. For example, the data acquisition and alignment module, topology inversion module, PV stripping and splicing module, and time-series deduction and self-correction module are used to implement steps S1, S2, S3, and S4 in the aforementioned multimodal data fusion PV power grid time-series simulation method, respectively. Therefore, its specific implementation can be referred to the description of the corresponding embodiments. To avoid redundancy, it will not be repeated here.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for time-series simulation of photovoltaic power station areas using multimodal data fusion, characterized in that, include: Acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area that includes static physical information, dynamic measurement information, and environmental information; Based on measurement data under no photovoltaic conditions at night, the equivalent topology of the transformer area is inverted using the voltage drop pattern. The equivalent topology includes at least the main line and the equivalent branches derived when the local power change and voltage fluctuation are mismatched. Under daytime conditions with sunlight, based on the environmental information in the multimodal time-series feature matrix of the transformer area, the photovoltaic power flow is separated from the net power at the user end to obtain independent photovoltaic injection objects and pure load objects, and the two are respectively connected to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology. Using the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as input, continuous time-series state deduction is performed to obtain the time-series simulation results of the voltage of each node and the power flow of each branch. Based on the residuals between the simulation results and the measured data, the equivalent branch impedance parameters are corrected online, and the self-corrected all-day time-series power flow deduction results are output.

2. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method for acquiring multi-source heterogeneous data within a transformer area and performing unified spatiotemporal alignment to form a multimodal time-series feature matrix of the transformer area containing static physical information, dynamic measurement information, and environmental information includes: Acquire static physical data, dynamic measurement data, and environmental time-series data. The static physical data includes the relationship between the user and the transformer, the incomplete topology of the GIS, and the unit impedance parameters of the conductors. The dynamic measurement data includes the user's active power, reactive power, and voltage time-series data. The environmental time-series data includes irradiance, weather conditions, and ambient temperature. Using user ID and sampling time as a unified index, the static physical data, dynamic measurement data and environmental time series data are mapped to the same user-time coordinate to obtain the basic fusion features of each user at each time. Extract the no-photovoltaic period at night based on irradiance, calculate the reactive power-voltage sensitivity of each user during the no-photovoltaic period at night, and fuse the basic fusion feature with the reactive power-voltage sensitivity to obtain the multimodal fusion feature vector of each user at each time. The multimodal fusion feature vectors of all users at the same time are arranged according to the user dimension to obtain the multimodal time series feature matrix of the transformer area at that time, and then arranged in chronological order to form a sequence of multimodal time series feature matrices for the whole day.

3. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method for retrieving the equivalent topology of a transformer substation based on measurement data under nighttime photovoltaic-free conditions and utilizing voltage drop patterns includes: Extract voltage data for each user during the nighttime period without photovoltaic power generation, calculate the nighttime average voltage value for each user based on the rated voltage of the low-voltage side of the distribution transformer, and select the user with the lowest nighttime average voltage value as the farthest user. Based on the nighttime voltage drop, active power, reactive power, and conductor unit impedance parameters of the farthest user, the equivalent length of the main line is calculated in reverse, and a single main line is constructed with the distribution transformer as the starting point and the farthest user as the ending point. Based on the relative relationship between the nighttime average voltage values ​​of the remaining users and the nighttime average voltage values ​​of the farthest user, the remaining users are mapped to their corresponding electrical locations on the single trunk line. When the power variation and voltage fluctuation of users in a local area do not match on the single trunk line, an equivalent branch is derived at the corresponding electrical location, and users in the local area are assigned to the equivalent branch, generating an equivalent topology of the transformer area that includes the trunk line and at least one equivalent branch.

4. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 3, characterized in that, When the power variation and voltage fluctuation of users in a local area do not match on the single trunk line, the method for deriving an equivalent branch at the corresponding electrical location includes: Calculate the correlation between the actual voltage timing of each user during the nighttime period without photovoltaic power and the estimated voltage timing at the mapped location on the main line, and include users with a correlation less than a set threshold into the branch derivation discrimination object. Local areas are divided according to the user's mapping location on the trunk line. The power throughput and average voltage fluctuation of each local area are calculated, and a power-voltage mismatch index is constructed. When the power-voltage mismatch index is greater than the set threshold, it is determined that there is a power throughput and voltage fluctuation mismatch in the local area, and an equivalent branch is derived at the average electrical location on the main line in the local area. The equivalent length of the derived equivalent branch is calculated based on the remaining voltage drop of each user in the local area, and the users in the local area are then separated from the trunk line and redistributed to the derived equivalent branch.

5. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method for separating photovoltaic power flow from net power at the user end to obtain independent photovoltaic injection objects and pure load objects based on environmental information in the multimodal time-series characteristic matrix of the transformer area under daytime lighting conditions includes: Extracting the daytime sunlight hours, and based on irradiance, ambient temperature, and photovoltaic installed capacity, calculating the theoretical photovoltaic active power output and photovoltaic reactive power output of each user during the daytime sunlight hours based on the photovoltaic inverter output mechanism model; Based on the net active power of each user during the daytime sunlight period and the theoretical photovoltaic active power output, the pure load active power demand after stripping the photovoltaic output is calculated, and the pure load reactive power demand is obtained by stripping the photovoltaic reactive power from the net reactive power. Using the topology nodes corresponding to each user as indexes, the extracted photovoltaic active power output and photovoltaic reactive power output are constructed into virtual photovoltaic injection objects, and the extracted pure load active power demand and pure load reactive power demand are constructed into pure load objects.

6. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 5, characterized in that, The method for connecting the two to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology includes: The virtual photovoltaic injection object and the pure load object are injected as dual-drive objects into the corresponding node ports of each user in the equivalent topology; Calculate the net injected active power and net injected reactive power of each user node based on the algebraic sum of the power of the virtual photovoltaic injection object and the pure load object at each user node. When the net injected active power is positive, it is determined that the user node injects active power into the distribution area; when the net injected active power is negative, it is determined that the user node absorbs active power from the distribution area. Using the equivalent topology as a spatial carrying skeleton, the virtual photovoltaic injection objects and pure load objects of all user nodes are attached to the corresponding nodes to form the photovoltaic-load collaborative simulation topology.

7. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method for obtaining the time-series simulation results of each node voltage and each branch power flow by performing continuous time-series state deduction using the photovoltaic-load co-simulation topology and the all-day multimodal time-series characteristic matrix as input includes: Based on the upstream and downstream relationships of each equivalent branch in the photovoltaic-load collaborative simulation topology, the active power flow and reactive power flow of each equivalent branch at each time point are calculated. Based on the impedance parameters of each equivalent branch and its active and reactive power flows, the voltage drop from the distribution transformer to each node is calculated, and the simulated voltage of each node at each moment is calculated with the rated voltage of the low-voltage side of the distribution transformer as the benchmark. Arrange the simulated node voltages at all times in chronological order to obtain the time-series simulation results of node voltages throughout the day. Arrange the active and reactive power flows of each branch at all times in chronological order to obtain the time-series simulation results of branch power flows throughout the day.

8. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method for online correction of equivalent branch impedance parameters based on the residuals of simulation results and measured data, and outputting self-corrected all-day time-series power flow projection results, includes: The simulated voltage of each node at each time point is compared with the measured voltage of the same node at the same time point. The voltage residual of each node at each time point is calculated, and an evaluation index for the voltage simulation error throughout the day is constructed based on the voltage residual of all nodes and all time points. With the goal of minimizing the evaluation index of the all-day voltage simulation, the partial derivatives of the evaluation index with respect to the resistance and reactance of each equivalent branch are calculated, and the resistance and reactance of each equivalent branch are updated according to the set correction step size. The updated equivalent branch impedance parameters are re-substituted into the photovoltaic-load co-simulation topology, and the continuous time-series state deduction and impedance parameter update are repeated until the convergence criterion is met. The self-corrected all-day time-series power flow deduction results are then output.

9. The photovoltaic substation time-series simulation method based on multimodal data fusion according to claim 1, characterized in that, The method is executed independently by local terminal equipment in the low-voltage distribution area. The local terminal equipment acquires dynamic measurement data collected by smart meters in the distribution area and environmental time-series data collected by locally deployed meteorological sensors, and completes the entire process calculation of topology inversion, photovoltaic stripping, time-series extrapolation and impedance self-correction locally, without relying on external servers or cloud platforms.

10. A photovoltaic substation time-series simulation system with multimodal data fusion, characterized in that, include: The data acquisition and alignment module is used to acquire multi-source heterogeneous data within the transformer area and perform unified spatiotemporal alignment to form a multimodal temporal feature matrix of the transformer area that includes static physical information, dynamic measurement information and environmental information. The topology inversion module is used to invert the equivalent topology of the transformer area based on measurement data under nighttime photovoltaic-free conditions and the voltage drop pattern. The equivalent topology includes at least the main line and the equivalent branches derived when the local power changes and voltage fluctuations are mismatched. The photovoltaic stripping and attaching module is used to strip the photovoltaic power flow from the net power of the user end under daytime sunlight conditions, based on the environmental information in the multimodal time-series feature matrix of the transformer area, to obtain independent photovoltaic injection objects and pure load objects, and attach the two to the corresponding nodes of the equivalent topology to form a photovoltaic-load collaborative simulation topology. The timing simulation and self-correction module is used to perform continuous timing state simulation with the photovoltaic-load co-simulation topology and the all-day multimodal timing feature matrix as input, to obtain the timing simulation results of the voltage of each node and the power flow of each branch, and to correct the equivalent branch impedance parameters online based on the residuals of the simulation results and the measured data, and output the self-corrected all-day timing power flow simulation results.

Images