Networking type energy storage inverter control method based on complex network space-time optimization

Abstract

The invention discloses a network construction type energy storage inverter control method based on complex network space-time optimization, and the method comprises the steps: positioning a historical timestamp on a time axis, constructing a function dependence graph of the historical timestamp, enabling the function dependence graph to comprise K nodes representing a power grid function unit, and enabling the K nodes to represent the power grid function unit according to a preset time step length, a function dependency graph of L historical timestamps is constructed in a sliding mode, a dependency graph sequence is generated on the basis of sequential arrangement of the historical timestamps, the dependency graph sequence is cut into M space-time dependency samples, iteration supervision training is conducted on a graph neural network model on the basis of the M space-time dependency samples, a space-time state prediction model is obtained, and a space-time state prediction model is obtained on the basis of the space-time state prediction model. According to the method, the space-time dependency degree is determined as the edge weight, the prediction capability of the model on the future state of the power grid is enhanced, and the network-building type energy storage inverter is supported to generate the prediction-based control instruction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of energy storage inverter control, in particular to a network construction type energy storage inverter control method based on complex network space-time optimization. BACKGROUND

[0002] With the increasing proportion of new energy such as photovoltaic power generation and wind power generation, the voltage and frequency fluctuation of the power grid is intensified, and the regulation difficulty of power balance is increased. The control method of the existing network construction type energy storage inverter can generate control instructions based on historical or real-time operation data; for example, the prior art disclosed in document No. CN119298603A discloses an energy storage inverter and a control method thereof, which can reduce the risk of device overcurrent damage by adjusting the output voltage to reduce the load current. However, such methods rely on local or local operating state information when generating control instructions, lack of perception of the overall space-time evolution trend of the system, resulting in insufficient foresight of the control instructions.

[0003] In addition, when the existing technology models the overall function unit of the power grid, it usually does not consider its geographical spatial distribution relationship. For example, there may be stronger voltage influence between photovoltaic units and load units that are close in distance, while the power response between distant units often has obvious delay, which cannot reflect the difference in spatial position and is difficult to accurately predict local voltage fluctuations.

[0004] At the same time, although some schemes use graph neural networks for state prediction, the graph structure constructed by these schemes relies on fixed electrical topology, and the edge weight between nodes is usually set as a constant or manually set, which cannot reflect the dynamic dependence relationship between the function units of the power grid that changes with the operating condition. The ability to capture the trend of system state changes is limited, making it difficult to support the network construction type energy storage inverter to generate highly reliable forward-looking control instructions under complex space-time conditions. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a network construction type energy storage inverter control method based on complex network space-time optimization, which solves the technical problems proposed in the background art by introducing edge weights based on time dependence and spatial dependence and constructing a function dependence graph based on the edge weights.

[0006] To achieve the above purpose, the present application is realized by the following technical scheme: A network construction type energy storage inverter control method based on complex network space-time optimization, comprising the following steps: S1, positioning a historical timestamp on a time axis and constructing a function dependence graph of the historical timestamp; wherein the function dependence graph comprises K nodes representing the function units of the power grid; S2, slide the functional dependency graph of L historical timestamps with a preset time step, and generate a dependency graph sequence based on the sequence arrangement of the historical timestamps; S3, cut the dependency graph sequence into M spatiotemporal dependency samples; S4, based on the M spatiotemporal dependency samples, iteratively supervise the training of the graph neural network model to obtain a spatiotemporal state prediction model; S5, based on the spatiotemporal state prediction model, determine the control instruction of the network-type energy storage inverter.

[0007] In some specific embodiments, the functional dependency graph of the historical timestamp is constructed, including: S1-1, obtaining a functional node set containing K nodes under the historical timestamp; S1-2, establishing a directed edge between the K nodes of the functional node set, S1-3, determining the edge weight based on the upstream and downstream power grid functional units on the directed edge; S1-4, traversing the functional node set, repeatedly establishing the directed edge and its edge weight until the functional dependency graph corresponding to the K power grid functional units is constructed.

[0008] In some specific embodiments, the functional node set containing K nodes under the historical timestamp is obtained, including: S1-1-1, delimiting a target collection area on the electronic map and anchoring the K power grid functional units in the target collection area; S1-1-2, obtaining a plurality of operating state parameters of the K power grid functional units at the same historical timestamp; S1-1-3, performing feature standardization on the plurality of operating state parameters to construct the operating state features corresponding to the power grid functional units; S1-1-4, defining the K power grid functional units as K nodes of a graph structure, and the corresponding operating state features as node vectors, to construct the functional node set; Wherein, the K nodes of the functional node set are all established with node indexes, and each node index corresponds to its power grid functional unit.

[0009] In some specific embodiments, the directed edge is established between the K nodes of the functional node set, including: S1-2-1, in the functional node set, anchoring a target node and selecting J candidate dependency nodes having candidate dependency relationship with the target node to obtain J candidate node pairs, wherein J=K-1; S1-2-2, for each candidate node pair, determining whether the candidate dependency node belongs to an effective dependency node based on a pre-constructed knowledge fact graph; S1-2-3, if it belongs to an effective dependency node, extract a predefined entity relationship in the knowledge fact graph as a dependency relationship between the target node and the effective dependency node; S1-2-4, based on the dependency relationship, establish a directed edge between the target node and the effective dependency node.

[0010] In some specific embodiments, determining whether a candidate dependency node in the pre-constructed knowledge fact graph belongs to an effective dependency node includes: S1-2-2-1, anchoring the two power grid functional units corresponding to the candidate node pair; S1-2-2-2, knowledge fact annotation of the two power grid functional units, and pairing to generate a to-be-judged entity pair; S1-2-2-3, in the pre-constructed knowledge fact graph, traversing each knowledge fact triple, extracting its head entity and tail entity, and pairing to generate several fact entity pairs; S1-2-2-4, calculating the matching degrees of the to-be-judged entity pair and the several fact entity pairs; S1-2-2-5, selecting the maximum matching degree from the several matching degrees, and comparing it with the set threshold; S1-2-2-6, if the maximum matching degree is greater than the set threshold, it is determined that the candidate dependency node in the candidate node pair belongs to an effective dependency node; otherwise, it belongs to an invalid dependency node.

[0011] In some specific embodiments, determining the edge weight based on the power grid functional units upstream and downstream of the directed edge includes: S1-3-1, calculating the spatial dependency degree of the directed edge; S1-3-2, calculating the time dependency degree of the directed edge; S1-3-3, weighting and fusing the spatial dependency degree and the time dependency degree to construct a spatio-temporal dependency degree, and determining the spatio-temporal dependency degree as the edge weight.

[0012] In some specific embodiments, calculating the spatial dependency degree of the directed edge includes: S1-3-1-1, anchoring the target node and the effective dependency node upstream and downstream of the directed edge, S1-3-1-2, extracting the two power grid functional units corresponding to the target node and the effective dependency node according to the node index of the target node and the effective dependency node; S1-3-1-3, obtaining the spatial coordinates of the two power grid functional units in the electronic map; S1-3-1-4, calculating the spatial distance between the two power grid functional units according to the spatial coordinates; S1-3-1-5, inversely proportional to the spatial distance between the target node and the effective dependency node. In some specific embodiments, the time dependence of the directed edge is calculated, comprising: S1-3-2-1, anchoring the historical timestamp on the time axis; S1-3-2-2, determining a fixed-length time window on the time axis based on the historical timestamp; S1-3-2-3, obtaining the running state parameters of the target node and the effective dependent node at several consecutive timestamps within the time window, and arranging the running state parameters based on the order of the several consecutive timestamps into a target parameter sequence and a dependent parameter sequence, respectively; S1-3-2-4, calculating the mean of the target parameter sequence and the dependent parameter sequence; S1-3-2-5, calculating the non-normalized covariance and non-normalized standard deviation of the target parameter sequence and the dependent parameter sequence according to the mean of the target parameter sequence and the dependent parameter sequence; S1-3-2-6, calculating the time dependence between the target node and the effective dependent node according to the non-normalized covariance and the non-normalized standard deviation of the target parameter sequence and the dependent parameter sequence; The calculation formula of the time dependence is: ; Wherein, denotes the time dependence, denotes the i-th target node, denotes the j-th effective dependent node, denotes the Pearson coefficient, denotes the absolute value of the Pearson coefficient; denotes the running state parameter of the i-th target node at the k-th consecutive timestamp, denotes the mean of the target parameter sequence, denotes the running state parameter of the j-th effective dependent node at the k-th consecutive timestamp, denotes the mean of the dependent parameter sequence, denotes the non-normalized covariance, denotes the non-normalized standard deviation product.

[0013] S1-3-2-7, obtaining the absolute value of the Pearson coefficient, defined as the time dependence.

[0014] In some specific embodiments, the dependency graph sequence is cut into M spatiotemporal dependence samples, comprising: S3-1, setting the preprocessed sample length; S3-2, based on the preprocessed sample length, sequentially cutting the dependency graph subsequence from the dependency graph sequence; The dependency graph subsequence comprises a functional dependency graph of consecutive historical time stamps. S3-3, for each dependency graph subsequence, the running state parameters of the K nodes in the functional dependency graph corresponding to the end time stamp are taken as target label values. S3-4, each dependency graph subsequence and the corresponding target label value are combined to form the spatio-temporal dependency sample.

[0015] In some specific embodiments, based on the spatio-temporal state prediction model, the control instruction of the grid-forming energy storage inverter is determined, comprising: S5-1, obtaining the running state parameters of the K grid function units in the target collection area at the current time stamp, and constructing the functional dependency graph of the current time stamp based on the running state parameters; S5-2, inputting the functional dependency graph of the current time stamp into the spatio-temporal state prediction model to output the predicted state parameters of the K grid function units at the target time stamp; S5-3, generating the control instruction of the grid-forming energy storage inverter according to the predicted state parameters of the K grid function units at the target time stamp.

[0016] The application provides a grid-forming energy storage inverter control method based on complex network spatio-temporal optimization, which has the following beneficial effects: The application can reflect the proximity relationship of the grid function units in geographical space by calculating the spatial distance between the upstream target node and the effective dependent node on the directed edge and inversely proportional to the spatial dependency between the target node and the effective dependent node, and introduce spatial distribution in the construction process of the functional dependency graph, so that the model has the perception ability of geographical layout when identifying the functional dependency relationship, which helps to distinguish the influence difference of the grid function units at different spatial positions on the running state parameters.

[0017] Further, the application uses the absolute value of the Pearson coefficient to calculate the time dependency, which can capture the dynamic response consistency of the grid function units on the running state parameters at consecutive time stamps, and can identify the grid function unit pairs with strong correlation in the time evolution process, and improve the representation of dynamic behavior.

[0018] Further, the application weights and fuses the spatial dependency and the time dependency into the spatio-temporal dependency as the edge weight, which considers the proximity of the grid function units in geographical space and also fuses the dynamic response consistency in the time sequence, so that the functional dependency graph can comprehensively reflect the coupling inside the grid.

[0019] Further, the application determines the space-time dependence as an edge weight, so that the directed edge in the functional dependence graph not only represents the connection direction, but also quantifies the relative strength of the dependence relationship; the strength is jointly determined by spatial proximity and temporal synchronicity, so that the graph structure can dynamically reflect the comprehensive influence degree between the power grid functional units in the historical period; provides higher quality sequence samples for the graph neural network model, enhances the prediction ability of the model for the future state of the power grid, and further supports the grid-forming energy storage inverter to generate prediction-based control instructions. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of the grid-forming energy storage inverter control method based on complex network space-time optimization of the application; Figure 2 A flowchart of the construction process of the functional dependence graph of the application; Figure 3 A flowchart of the determination process of the effective dependence node of the application; Figure 4 A flowchart of the calculation process of the spatial dependence of the application; Figure 5 A flowchart of the calculation process of the temporal dependence of the application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0022] Embodiment 1: Please refer to Figure 1 The application provides a grid-forming energy storage inverter control method based on complex network space-time optimization, which comprises the following application steps: S1, positioning a historical timestamp on a time axis, and constructing a functional dependence graph of the historical timestamp; wherein the functional dependence graph comprises K nodes representing power grid functional units; In this embodiment, the power grid functional units represent the units corresponding to the wind, light, diesel and storage of the power grid system; exemplary, the K power grid functional units include: a photovoltaic power generation unit, a wind power generation unit, a diesel power generation unit, a storage unit, a load cluster unit and a power grid connection unit.

[0023] S2, slidingly constructing functional dependence graphs of L historical timestamps at a preset time step, and generating a dependence graph sequence based on the order arrangement of the historical timestamps; S3, cutting the dependence graph sequence into M space-time dependence samples; S4, based on the M spatio-temporal dependent samples, iteratively supervising training of the graph neural network model to obtain a spatio-temporal state prediction model; Specifically, the iterative training step of the spatio-temporal state prediction model comprises: sequentially inputting the M spatio-temporal dependent samples into the graph neural network model; The model outputs a preliminary prediction result of the future operating state of the power grid functional unit according to the dependent graph sequence in the input sample; Comparing the preliminary prediction result with the actually collected operating state parameters to calculate a prediction error; Based on the prediction error, reversely adjusting internal parameters of the graph neural network model; Repeating the above process until the prediction result output by the model converges or reaches a preset training accuracy, completing the model training.

[0024] S5, determining a control instruction of the grid-forming energy storage inverter based on the spatio-temporal state prediction model.

[0025] In the embodiment, by constructing a functional dependency graph representing the power grid functional unit and its dependency relationship under the historical timestamp, a dependent graph sequence arranged in time sequence is generated, which is cut into spatio-temporal dependent samples for iterative supervised training of the graph neural network model to generate a spatio-temporal state prediction model, and based on the model, a control instruction of the grid-forming energy storage inverter is determined, realizing spatio-temporal modeling and prediction driving from the historical operating state of the power grid functional unit to the future control behavior, thereby improving the prediction accuracy of the power grid voltage, frequency and power change trend, supporting the grid-forming energy storage inverter to generate a control instruction with grid support role in advance under complex operating conditions, and enhancing the active adjustment capability of the power grid operation.

[0026] Embodiment 2: refer to Figures 2 to 5 The technical solution of the embodiment 2 differs from that of the embodiment 1 in that the embodiment 1 discloses specific execution sub-steps of each application step.

[0027] The specific execution sub-steps of the control method step S1 are: S1-1, obtaining a functional node set containing K nodes under the historical timestamp; S1-2, establishing a directed edge between the K nodes of the functional node set, S1-3, determining an edge weight based on the upstream and downstream power grid functional units of the directed edge; S1-4, traversing the functional node set, repeatedly establishing the directed edge and its edge weight until the functional dependency graph corresponding to the K power grid functional units is constructed.

[0028] In the embodiment, the constructed function dependency graph realizes the structural expression of the connection direction and interaction intensity of each power grid functional unit at the historical timestamp through the function node set containing K nodes, the directed edges between the nodes, and the edge weight determined based on the upstream and downstream power grid functional units.

[0029] Further, the sub-step S1-1 specifically includes: S1-1-1, delimiting a target collection area on an electronic map and anchoring K power grid functional units in the target collection area; S1-1-2, obtaining a plurality of operating state parameters of the K power grid functional units at the same historical timestamp; In the embodiment, the operating state parameter represents the core operating state parameter of the power grid functional unit at the timestamp, including the environmental input, the operating state, and the power response dimension.

[0030] Specifically, the operating state parameter of the photovoltaic power generation unit is the light intensity, the component temperature, and the output power; The operating state parameter of the wind power generation unit is the wind speed, the component rotating speed, and the output power; The operating state parameter of the diesel power generation unit is the fuel reserve, the unit oil consumption, and the output power; The operating state parameter of the energy storage unit is the state of charge, the terminal voltage, and the charge and discharge power; The operating state parameter of the load cluster unit is the load type (industrial / commercial / residential), the power consumption period, and the load power; The operating state parameter of the power grid connection unit is the grid connection point voltage, the grid connection point frequency, and the electricity price period (peak / flat / valley); S1-1-3, performing feature standardization on the plurality of operating state parameters to construct the operating state feature corresponding to the power grid functional unit; Specifically, the feature standardization represents the dimensionless normalization processing on each operating state parameter to eliminate the numerical difference between different physical quantities; the feature standardization adopts the Z-score standardization or the Min-Max normalization method; And the operating state feature represents the dimensionless feature vector obtained after the standardization processing and used for representing the operating state of the power grid functional unit.

[0031] S1-1-4, defining the K power grid functional units as K nodes of the graph structure, the corresponding operating state features as node vectors, and constructing a function node set; Among them, the K nodes of the function node set are established with node indexes, and each node index corresponds to its power grid functional unit.

[0032] In the embodiment, the constructed functional node set is obtained by mapping K power grid functional units in a target collection area as K nodes in a graph structure, and using the standardized operating state features as the node vectors of the nodes, so that the operating state parameters of the power grid functional units at the same historical timestamp are organized as node indexes, and a unified data basis is provided for establishing directed edges between the nodes of the functional node set and determining the edge weights.

[0033] Further, the sub-step S1-2 specifically comprises: S1-2-1, in the functional node set, anchoring a target node, and selecting J candidate dependent nodes having candidate dependent relationships with the target node, to obtain J candidate node pairs, wherein J=K-1; Specifically, in the functional node set, the candidate dependent node refers to the other nodes corresponding to the remaining K-1 power grid functional units except the target node. It may have potential interaction possibilities with the target node in electrical topology, geographical proximity or historical operation data.

[0034] S1-2-2, for each candidate node pair, determining whether the candidate dependent node belongs to an effective dependent node based on the pre-constructed knowledge fact graph; The determination process of the effective dependent node is as follows: S1-2-2-1, anchoring the two power grid functional units corresponding to the candidate node pair; S1-2-2-2, knowledge fact labeling is performed on the two power grid functional units, and a to-be-determined entity pair is generated by pairing; S1-2-2-3, in the pre-constructed knowledge fact graph, traversing each knowledge fact triple, extracting the head entity and the tail entity, and pairing to generate a plurality of fact entity pairs; S1-2-2-4, calculating a plurality of matching degrees of the to-be-determined entity pair and the plurality of fact entity pairs; Exemplarily, the matching degree is determined by mapping the power grid functional unit names in the to-be-determined entity pair and the fact entity pair into pre-trained power field word vectors, and calculating the cosine similarity between the corresponding word vectors; The matching degree of the to-be-determined entity pair and a certain fact entity pair is the average of the head entity word vector similarity and the tail entity word vector similarity.

[0035] S1-2-2-5, selecting the maximum matching degree from the plurality of matching degrees, and comparing it with a set threshold; S1-2-2-6, if the maximum matching degree is greater than the set threshold, it is determined that the candidate dependent node in the candidate node pair belongs to an effective dependent node; otherwise, it belongs to an invalid dependent node.

[0036] Specifically, the knowledge fact graph is a predefined rule-based knowledge base, which contains the physical connection rules and operation coupling relationships between various types of power grid functional units in the power grid, for example: (Photo-voltaic array, power supply, load cluster) (energy storage system, support, power grid frequency) (diesel generator, backup power supply, critical load) (wind turbine, controlled by, wind speed) Each knowledge fact is in the form of a triple: (head entity, relationship, tail entity), which constitutes the knowledge fact triple.

[0037] In this embodiment, by anchoring the corresponding two power grid functional units of the candidate node pair, the knowledge fact annotation generates the entity pair to be determined, the knowledge fact triples in the knowledge fact graph are traversed to generate the fact entity pair, the matching degree between the entity pair to be determined and the fact entity pair is calculated, the maximum matching degree is selected and compared with the set threshold, and whether the candidate dependent node belongs to the valid dependent node is determined according to the comparison result, and the validity of the dependency relationship between the power grid functional units is confirmed.

[0038] Further, the sub-step S1-2 specifically further includes S1-2-3, if it belongs to the valid dependent node, extracting the predefined entity relationship in the knowledge fact graph as the dependency relationship between the target node and the valid dependent node; S1-2-4, based on the dependency relationship, establishing a directed edge between the target node and the valid dependent node.

[0039] In this embodiment, by anchoring the target node between the K nodes of the functional node set and selecting the candidate dependent node, the validity of the candidate dependent node is determined in combination with the pre-constructed knowledge fact graph, the dependency relationship is determined according to the entity relationship in the knowledge fact graph, and the directed edge is established based on the dependency relationship, thereby realizing the establishment of the connection direction between the power grid functional units.

[0040] Further, the sub-step S1-3 specifically includes: S1-3-1, calculating the spatial dependency degree of the directed edge; S1-3-2, calculating the time dependency degree of the directed edge; S1-3-3, weighting and fusing the spatial dependency degree and the time dependency degree to construct the space-time dependency degree, and determining the space-time dependency degree as the edge weight.

[0041] In this embodiment, by calculating the spatial dependency degree and the time dependency degree of the directed edge respectively, and weighting and fusing them into the edge weight, the calculation results of the spatial dependency degree and the time dependency degree are integrated into the edge weight, thereby completing the quantitative establishment of the connection strength between the power grid functional units.

[0042] Exemplarily, the calculation of the spatial dependency comprises: S1-3-1-1, anchoring the target node and the effective dependency node on the upstream and downstream of the directed edge, S1-3-1-2, extracting the corresponding two power grid functional units according to the node indexes of the target node and the effective dependency node; S1-3-1-3, obtaining the spatial coordinates of the two power grid functional units on the electronic map; S1-3-1-4, calculating the spatial distance between the two power grid functional units according to the spatial coordinates; Exemplarily, the spatial distance can be calculated based on the Euclidean distance between the two spatial coordinates.

[0043] S1-3-1-5, inversely calculating the spatial dependency between the target node and the effective dependency node according to the spatial distance; Exemplarily, the inverse calculation formula of the spatial dependency is: ; wherein, represents the spatial dependency, represents the spatial distance between the ith target node and the jth effective dependency node, is the attenuation coefficient of the spatial distance; in the embodiment, the role of the attenuation coefficient is to adjust the nonlinear influence intensity of the spatial distance on the spatial dependency. Specifically: When the attenuation coefficient increases, the influence of the spatial distance on the spatial dependency will be more significant. That is, as the spatial distance increases, the weight decreases faster.

[0044] When the attenuation coefficient decreases, the influence of the spatial distance on the spatial dependency will weaken. That is, as the spatial distance increases, the weight decreases slower.

[0045] In the embodiment, by anchoring the target node and the effective dependency node on the upstream and downstream of the directed edge, extracting the corresponding power grid functional units according to the node indexes, obtaining the spatial coordinates of the power grid functional units on the electronic map, calculating the spatial distance, and inversely calculating the spatial dependency according to the spatial distance, the expression of the spatial relationship between the power grid functional units is completed.

[0046] Exemplarily, the calculation process of the time dependency comprises: S1-3-2-1, anchoring the historical timestamp on the time axis; S1-3-2-2, determining a fixed-length time window on the time axis based on the historical timestamp; Exemplarily, in the embodiment, the time window is , represents that the time sub-axis with length T is intercepted as a time window by backtracking T length at the historical timestamp t; wherein, the length T is preferably 1 hour.

[0047] S1-3-2-3, within the time window, the running state parameters of the target node and the effective dependent nodes at several continuous timestamps are acquired, and the running state parameters are arranged into a target parameter sequence and a dependent parameter sequence based on the order of the several continuous timestamps respectively; Exemplarily, the running state parameter sequence represents the core dynamic variables of the node, such as: photovoltaic power generation unit→output power; energy storage unit→charge and discharge power; load cluster unit→load power; power grid-connected unit→grid-connected point evaluation rate; Of course, the core dynamic variable only needs to be able to reflect the energy interaction behavior of the power grid functional unit, and can be selected based on the actual system demand.

[0048] S1-3-2-4, the mean values of the target parameter sequence and the dependent parameter sequence are calculated; S1-3-2-5, according to the mean values of the target parameter sequence and the dependent parameter sequence, the unnormalized covariance and the unnormalized standard deviation of the target parameter sequence and the dependent parameter sequence are calculated; S1-3-2-6, according to the unnormalized covariance and the unnormalized standard deviation of the target parameter sequence and the dependent parameter sequence, the time dependence between the target node and the effective dependent nodes is calculated; The calculation formula of the time dependence is: ; wherein, represents the time dependence, represents the i-th target node, represents the j-th effective dependent node, represents the Pearson coefficient, represents the absolute value of the Pearson coefficient; represents the running state parameter of the i-th target node at the k-th continuous timestamp, represents the mean value of the target parameter sequence, represents the running state parameter of the j-th effective dependent node at the k-th continuous timestamp, represents the mean value of the dependent parameter sequence, represents the unnormalized covariance, represents the unnormalized standard deviation product.

[0049] S1-3-2-7, obtain the absolute value of the Pearson coefficient, defined as the time dependence.

[0050] In the embodiment, by determining a fixed-length time window based on a historical timestamp, obtaining the running state parameters of the target node and the effective dependent nodes at several consecutive timestamps in the time window, and constructing the target parameter sequence and the dependent parameter sequence, the mean, the non-normalized covariance and the non-normalized standard deviation of the two sequences are calculated, the correlation is calculated based on the Pearson coefficient formula, and the absolute value is taken as the time dependence, thereby completing the quantitative representation of the change relationship of the running state parameters of the power grid functional units at consecutive timestamps.

[0051] In the embodiment, the specific execution sub-steps of the control method step S3 are: S3-1, set a pre-processed sample length; S3-2, based on the pre-processed sample length, sequentially cut dependent graph subsequences from the dependent graph sequence; The dependent graph subsequence contains functional dependency graphs of consecutive historical timestamps. S3-3, for each dependent graph subsequence, the running state parameters of the K nodes in the functional dependency graph corresponding to the end timestamp are taken as the target label value; S3-4, combine each dependent graph subsequence and its corresponding target label value to form the spatiotemporal dependent sample.

[0052] In the embodiment, by setting a pre-processed sample length, dependent graph subsequences containing functional dependency graphs of consecutive historical timestamps are sequentially cut from the dependent graph sequence, the running state parameters of the K nodes in the functional dependency graph corresponding to the end timestamp of each dependent graph subsequence are taken as the target label value, and the dependent graph subsequence and its corresponding target label value are combined to obtain the spatiotemporal dependent sample that can be used for model input.

[0053] In the embodiment, the specific execution sub-steps of the control method step S5 are: S5-1, obtain the running state parameters of the K power grid functional units in the target collection area at the current timestamp, and construct a functional dependency graph of the current timestamp based on the running state parameters; In the embodiment, the functional dependency graph of the current timestamp is constructed in the same method as the historical timestamp, including: obtaining the current running state parameters, constructing a functional node set, and establishing directed edges and edge weights based on a knowledge fact graph.

[0054] S5-2, input the functional dependency graph of the current timestamp into the spatiotemporal state prediction model to output the predicted state parameters of the K power grid functional units at the target timestamp; The predicted state parameter in the application refers to a parameter representing the running state of the power grid functional unit at a future target timestamp, including but not limited to: output power, terminal voltage, state of charge, grid-connected point frequency, environmental input, etc. Therefore, the running state parameter in the embodiment can be divided into "historical running state parameter", "current running state parameter" and "predicted state parameter" according to the time stage it is in, and the data ontology is consistent, only the time attribute is different.

[0055] S5-3, generating a control instruction of the grid-forming energy storage inverter according to the predicted state parameters of the K power grid functional units at the target timestamp; Specifically, the generation of the control instruction includes: If the comparison of the predicted state parameters finds that the grid-connected point frequency is lower than a set threshold, a control instruction for actively supporting the grid-connected point frequency is generated, so that the energy storage unit enters a discharging state and enhances the grid inertia response; If the comparison of the predicted state parameters finds that the load power will rise sharply and the photovoltaic / wind power output is insufficient, a control instruction for maintaining the bus voltage stability is generated, so that the energy storage unit releases active power in advance to provide voltage support; If the comparison of the predicted state parameters finds that the price period is about to enter the "valley period" and the state of charge is lower than a set threshold, a control instruction for low-cost energy storage is generated, so that the energy storage unit switches to a charging mode to absorb low-cost electricity; If the comparison of the predicted state parameters finds that the wind and light output is significantly higher than the load demand and the energy storage is not full, a control instruction for peak clipping and valley filling is generated, so that the energy storage unit absorbs excess electricity to balance the system power; The above control instructions include active / reactive power set values, virtual inertia, droop coefficients, and operation mode signals, which are dynamically configured according to the control architecture of the grid-forming inverter and delivered to the grid-forming energy storage inverter to drive it to complete the operation mode switching before the target timestamp.

[0056] In summary, in the embodiment, the connection direction and dependence strength between the power grid functional units are expressed by constructing a functional dependence graph, the evolution process of the system state at consecutive timestamps is reflected by using space-time dependence samples, the running state parameters at the target timestamp are generated based on the prediction model, and the corresponding control instructions are generated, so that the generation of the control instruction is extended from the running state at a single time point to a composite representation covering the spatial connection relationship and time evolution relationship between the power grid functional units, the control behavior changes from responsive adjustment to advance intervention based on prediction, the forward-looking regulation of the power grid voltage, frequency and power balance is realized, and the active adjustment capability of the grid-forming energy storage inverter in the complex network environment is supported.

[0057] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (for example, infrared, wireless, microwave, etc.) mode.

[0058] The above describes only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A network-constructed energy storage inverter control method based on complex network space-time optimization, characterized in that, The method comprises the following steps: Positioning a historical timestamp on a time axis and constructing a function dependency graph of the historical timestamp; wherein the function dependency graph contains K nodes representing power grid functional units; Slidingly constructing function dependency graphs of L historical timestamps with a preset time step, and generating a dependency graph sequence based on the sequential arrangement of the historical timestamps; Cutting the dependency graph sequence into M spatiotemporal dependency samples; Based on the M spatiotemporal dependency samples, iteratively supervising and training a graph neural network model to obtain a spatiotemporal state prediction model; Based on the spatiotemporal state prediction model, determining the control instruction of the grid-type energy storage inverter.

2. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 1, characterized in that, The method for constructing the function dependency graph of the historical timestamp comprises the following steps: Obtaining a function node set containing K nodes under the historical timestamp; Establishing a directed edge between the K nodes of the function node set, Determining the edge weight based on the upstream and downstream power grid functional units of the directed edge; Traversing the function node set, repeatedly establishing the directed edge and its edge weight until the function dependency graph corresponding to the K power grid functional units is constructed.

3. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 2, characterized in that, The method for obtaining the function node set containing K nodes under the historical timestamp comprises the following steps: Delimiting a target collection area on an electronic map and anchoring K power grid functional units within the target collection area; Obtaining a plurality of operating state parameters of the K power grid functional units at the same historical timestamp; Standardizing the plurality of operating state parameters to construct operating state features corresponding to the power grid functional units; Defining the K power grid functional units as K nodes of a graph structure, and the corresponding operating state features as node vectors to construct the function node set; wherein the K nodes of the function node set are all established with node indexes, and each node index corresponds to its power grid functional unit.

4. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 2, characterized in that, The method for establishing a directed edge between the K nodes of the function node set comprises the following steps: In the function node set, anchoring a target node and selecting J candidate dependency nodes having candidate dependency relationships with the target node to obtain J candidate node pairs, wherein J=K-1; For each candidate node pair, determining whether the candidate dependency node therein belongs to an effective dependency node based on a pre-constructed knowledge fact graph; If it belongs to an effective dependency node, extracting a pre-defined entity relationship in the knowledge fact graph as the dependency relationship between the target node and the effective dependency node; Based on the dependency relationship, establishing a directed edge between the target node and the effective dependency node.

5. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 4, characterized in that, The method for determining whether the candidate dependency node belongs to an effective dependency node based on the pre-constructed knowledge fact graph comprises the following steps: Anchoring the two power grid functional units corresponding to the candidate node pair; Performing knowledge fact annotation on the two power grid functional units to generate a pair of entities to be determined; In the pre-constructed knowledge fact graph, traversing each knowledge fact triple to extract its head entity and tail entity to generate a plurality of fact entity pairs; Calculating a plurality of matching degrees of the pair of entities to be determined and the plurality of fact entity pairs; Selecting the maximum matching degree from the plurality of matching degrees and comparing it with a set threshold; If the maximum matching degree is greater than the set threshold, it is determined that the candidate dependency node in the candidate node pair belongs to an effective dependency node; otherwise, it belongs to an ineffective dependency node.

6. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 2, characterized in that, The method for determining the edge weight based on the upstream and downstream power grid functional units of the directed edge comprises the following steps: Calculating the spatial dependency degree of the directed edge; calculating a time dependency of the directed edge; fusing the spatial dependency and the time dependency by weighting, constructing a space-time dependency, and determining the space-time dependency as the edge weight.

7. The complex network space-time optimization-based network-constructed energy storage inverter control method according to claim 6, characterized in that, calculating a spatial dependency of the directed edge, comprising: anchoring a target node and an effective dependency node downstream of the directed edge, extracting two power grid functional units corresponding to the target node and the effective dependency node according to node indexes of the target node and the effective dependency node; obtaining spatial coordinates of the two power grid functional units in the electronic map; calculating a spatial distance between the two power grid functional units according to the spatial coordinates; calculating the spatial dependency between the target node and the effective dependency node inversely proportional to the spatial distance.

8. The complex network space-time optimization-based network-constructed energy storage inverter control method according to claim 6, characterized in that, calculating a time dependency of the directed edge, comprising: anchoring the historical timestamp on a time axis; determining a fixed-length time window on the time axis based on the historical timestamp; obtaining running state parameters of the target node and the effective dependency node at a plurality of continuous timestamps within the time window, and arranging the running state parameters into a target parameter sequence and a dependency parameter sequence based on the order of the plurality of continuous timestamps respectively; calculating mean values of the target parameter sequence and the dependency parameter sequence; calculating an unnormalized covariance and an unnormalized standard deviation of the target parameter sequence and the dependency parameter sequence according to the mean values of the target parameter sequence and the dependency parameter sequence; calculating the time dependency between the target node and the effective dependency node according to the unnormalized covariance and the unnormalized standard deviation of the target parameter sequence and the dependency parameter sequence. The calculation formula of the time dependency is: ； wherein, denotes a time dependency, denotes the i-th target node, denotes the j-th valid dependency node, denotes a Pearson coefficient, denotes an absolute value of a Pearson coefficient; denotes the running state parameter of the i-th target node at the k-th consecutive time stamp, denotes a mean of the target parameter sequence, denotes the running state parameter of the j-th valid dependency node at the k-th consecutive time stamp, denotes a mean of the dependency parameter sequence, denotes an unnormalized covariance, denotes an unnormalized standard deviation product; obtaining an absolute value of the Pearson coefficient, defined as the time dependency.

9. The complex network space-time optimization-based network-structured energy storage inverter control method according to claim 1, characterized in that, cutting the dependency graph sequence into M space-time dependency samples, comprising: setting a preprocessed sample length; based on the preprocessed sample length, sequentially cutting dependency graph subsequences from the dependency graph sequence; wherein the dependency graph subsequence contains a functional dependency graph of continuous historical timestamps; for each dependency graph subsequence, taking running state parameters of K nodes in the functional dependency graph corresponding to the end timestamp of the dependency graph subsequence as target label values; combining each dependency graph subsequence and its corresponding target label values to form the space-time dependency sample.

10. The complex network space-time optimization-based network-constructed energy storage inverter control method according to claim 9, characterized in that, determining a control instruction of the grid-forming energy storage inverter based on the space-time state prediction model, comprising: obtaining running state parameters of K power grid functional units in a target collection area at a current timestamp, and constructing a functional dependency graph of the current timestamp based on the running state parameters; inputting the functional dependency graph of the current timestamp into the space-time state prediction model to output predicted state parameters of K power grid functional units at a target timestamp; generating a control instruction of the grid-forming energy storage inverter according to the predicted state parameters of K power grid functional units at the target timestamp.

Images