A multi-device cooperative energy efficiency optimization method and system for a prepackaged substation

Abstract

The application discloses a kind of prepackaged substation multi-device collaborative energy efficiency optimization method and system, it is related to power system energy efficiency optimization technical field, including: acquisition multi-device real-time load and energy storage level data, generate pressure distribution data after peak prediction;Estimate energy storage capacity and peak demand difference generates initial resource scheduling scheme;Initial resource scheduling scheme is input to equipment response simulation model and obtains operation bottleneck list;Collaborative interaction record in operation bottleneck list is identified to construct global dependency mapping for missing item;According to this dynamic weight adjustment obtains resource allocation coordination path;Resource allocation coordination path is checked and energy efficiency optimization configuration is obtained according to global dependency mapping adjustment;According to energy efficiency optimization configuration, adjust substation operating state and collect feedback data closed loop update.The application combines multi-device local decision and global dependency mapping by distributed control mode, relieves collaborative response delay and resource allocation uneven, improves peak period carrying capacity.

Description

Technical Field

[0001] This application relates to the field of power system energy efficiency optimization technology, specifically a multi-equipment collaborative energy efficiency optimization method and system for prefabricated substations. Background Technology

[0002] Prefabricated substations, as key equipment combinations responsible for power distribution and conversion in power distribution networks, integrate various electrical devices such as transformers, circuit breakers, energy storage units, and distribution automation terminals. With the increasing proportion of distributed power sources and the diversified development of electricity loads, prefabricated substations need to withstand greater power fluctuations without increasing land area or significant investment.

[0003] Existing energy efficiency optimization schemes for prefabricated substations mostly focus on the operation and adjustment of single devices, such as optimizing only the tap position of transformers or adjusting only the charging and discharging strategies of energy storage units. In scenarios where multiple devices operate simultaneously, these single-device-centric optimization schemes are prone to the following shortcomings: First, due to the lack of a global description of the mutual influence between multiple devices, when the resource demand of multiple devices increases simultaneously at a certain time, resource allocation is prone to contention, with some devices experiencing localized overload while the capacity of other devices is not fully utilized. Second, there are differences in response latency among multiple devices; relying solely on a single central controller for setting makes it difficult to identify which devices are experiencing response bottlenecks in a timely manner, causing the actual execution of dispatch commands to deviate from expectations. Third, the lack of a closed-loop update mechanism that combines historical load patterns with real-time feedback results in insufficient long-term adaptability of the optimization scheme to load distribution.

[0004] The aforementioned shortcomings of existing solutions collectively indicate that the key issues for improving the overall energy efficiency of prefabricated substations are how to identify collaborative response bottlenecks, construct a global mapping that reflects the dependencies between multiple devices, and dynamically adjust resource allocation accordingly in scenarios where multiple devices operate in parallel. Summary of the Invention

[0005] To address the above issues, this application provides a multi-equipment collaborative energy efficiency optimization method and system for prefabricated substations, which solves the problem that existing optimization schemes centered on a single equipment are not suitable for scenarios where multiple equipment operate simultaneously.

[0006] To achieve the above objectives, the technical solution adopted in this application is as follows: According to a first aspect of this application, a method for multi-equipment collaborative energy efficiency optimization in a prefabricated substation is provided, the method comprising: S1. Collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak time periods, and generate pressure distribution data based on the energy storage level data in each peak time period. S2, based on the pressure distribution data, evaluate the difference between the energy storage capacity and the demand during each peak period. If the difference exceeds the preset capacity threshold, generate an initial resource scheduling scheme according to the peak period. S3, input the initial resource scheduling scheme into the device response simulation model to obtain the response delay sequence of each device during the peak period, and record the devices whose response delay exceeds the preset delay threshold in the operation bottleneck list; S4, identify missing items in the collaborative interaction records of each device in the bottleneck list, construct a global dependency mapping between multiple devices based on the missing items, and obtain the device priority adjustment direction from the global dependency mapping; S5. Based on the device priority adjustment direction, dynamically adjust the resource allocation ratio in the initial resource scheduling scheme to obtain the resource allocation coordination path; S6, verify the resource allocation coordination path, and adjust the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs, to obtain an energy efficiency optimized configuration; S7, adjust the substation operation status according to the energy efficiency optimization configuration, collect operation feedback data and send it back to step S1 to update the peak prediction.

[0007] Preferably, the step of performing peak prediction on the real-time load data to divide it into multiple peak time periods includes: The real-time load data is filtered by sliding window mean filtering to obtain the filtered load sequence. The time series predictor trained based on historical load data is extrapolated to the filtered load sequence to obtain the predicted load sequence. The continuous time period in the predicted load sequence that exceeds the preset peak threshold is determined as the peak period.

[0008] Preferably, generating the initial resource scheduling scheme according to the peak time period includes: Based on the difference between each peak period, the peak periods are sorted from largest to smallest to obtain a time period priority sequence; Resource allocation is performed sequentially according to the time period priority sequence, starting with energy storage devices, load migration in adjacent time periods, and external power supply, until the difference is covered, thus obtaining the initial resource scheduling scheme.

[0009] Preferably, the step of identifying missing items in the collaborative interaction records of each device in the bottleneck list and constructing a global dependency mapping between multiple devices based on the missing items includes: Organize the collaborative interaction records of each device in the bottleneck list into a task transfer matrix, and mark the devices corresponding to rows or columns in the task transfer matrix where the number of missing items exceeds a preset missing threshold as decision blank nodes. Using each device in the substation as a node and the elements of the task transfer matrix between devices as weighted directed edges, a dependency directed graph is constructed. The decision blank nodes are merged into the directed graph of dependencies, and topological sorting and in-degree statistics are performed on the directed graph of dependencies to obtain the global dependency mapping.

[0010] Preferably, the global dependency mapping includes dependency scores for each device.

[0011] Preferably, the step of dynamically adjusting the resource allocation ratio in the initial resource scheduling scheme according to the device priority adjustment direction to obtain the resource allocation coordination path includes: Based on the dependency scores and bottleneck weights of each device, a comprehensive priority score is calculated; The resource allocation ratios of each device in the initial resource scheduling scheme are redistributed according to the comprehensive priority score from high to low, and the sum of the resource allocation ratios of all devices after redistribution is equal to 1, thus obtaining the resource allocation coordination path.

[0012] Preferably, the step of verifying the resource allocation coordination path and adjusting the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs includes: The resource allocation and coordination path is simulated a second time in the device response simulation model to obtain the load rate of each device. When the absolute value of the deviation between the load rate of at least one device and the average load rate of the device group exceeds a preset deviation threshold, it is determined that there is a local resource allocation deviation, and the device operating parameters are adjusted according to the associated devices of the at least one device in the global dependency mapping.

[0013] Preferably, the step of sending the collected operational feedback data back to step S1 to update the peak prediction includes: The voltage, current, and temperature data of the substation's operating status are collected according to a preset collection cycle as the operating feedback data; The operational feedback data is incorporated into the historical load database, so that the peak prediction prediction model is used to calculate the peak in the next scheduling cycle using the updated historical load data that includes the operational feedback data.

[0014] According to a second aspect of this application, a multi-equipment collaborative energy efficiency optimization system for a prefabricated substation employing the above-described multi-equipment collaborative energy efficiency optimization method for prefabricated substations is provided, the system comprising: The load prediction module is used to collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak periods, and generate pressure distribution data based on the energy storage level data in each peak period. The scheduling generation module is used to evaluate the difference between the energy storage capacity and the demand during each peak period based on the pressure distribution data, and generate an initial resource scheduling scheme according to the peak period when the difference exceeds a preset capacity threshold. The simulation test module is used to input the initial resource scheduling scheme into the device response simulation model, obtain the response delay sequence of each device, and record the devices whose response delay exceeds the preset delay threshold into the operation bottleneck list. The dependency building module is used to identify missing items in the collaborative interaction records of each device in the bottleneck list, build a global dependency mapping, and output the device priority adjustment direction. The weight adjustment module is used to dynamically adjust the resource allocation ratio in the initial resource scheduling scheme according to the device priority adjustment direction, so as to obtain the resource allocation coordination path. The configuration determination module is used to verify the resource allocation coordination path, and adjust the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs, so as to obtain an energy efficiency optimized configuration. The feedback update module is used to adjust the substation's operating status according to the energy efficiency optimization configuration and collect operating feedback data to send back to the load prediction module.

[0015] Preferably, in the feedback update module, voltage, current, and temperature data of the substation's operating status are collected according to a preset collection cycle as the operating feedback data, and the operating feedback data is incorporated into the historical load database for updating.

[0016] The embodiments of this application have the following advantages: By combining the local decision-making behavior of multiple devices with global dependency mapping, compared with optimization schemes that only focus on a single device, this application can take into account the priority and load-bearing capacity of each device when multiple devices are running simultaneously, thus alleviating resource contention and local overload.

[0017] By employing a two-stage process of first identifying the list of operational bottlenecks and then constructing dependency mappings, this application can proactively expose devices with response latency exceeding anomalies, avoiding the lag problem of traditional centralized controllers that only passively adjust after a bottleneck occurs, thereby improving the substation's carrying capacity during peak hours.

[0018] By sending operational feedback data back to the peak prediction stage and incorporating it into the historical database, the optimization scheme of this application has closed-loop update capability, adaptive capability to long-term changes in load distribution, and is conducive to load balancing under long-term operation.

[0019] The method of this application is suitable for deployment in a distributed control architecture. A single distributed controller only needs to rely on local data and the interaction records of a small number of neighboring devices to participate in the construction of global dependency mapping, which is convenient to implement in the compact space of a prefabricated substation. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of a multi-equipment collaborative energy efficiency optimization method for a prefabricated substation provided in an embodiment of this application; Figure 2 This is a structural block diagram of a multi-equipment collaborative energy efficiency optimization system for a prefabricated substation provided in an embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0022] The multi-device collaborative energy efficiency optimization method for prefabricated substations provided in this application embodiment is implemented using a distributed control architecture. Specifically, the substation distributed controller is located inside the prefabricated substation and is connected to the local control units of multiple devices such as transformers, energy storage units, circuit breakers, and feeder monitoring terminals within the substation via industrial Ethernet communication. This distributed controller does not need to upload all raw data to a higher-level station for centralized processing; instead, it completes the calculations from steps S1 to S7 locally and exchanges necessary collaborative interaction records with only a small number of neighboring control nodes, thus forming a distributed control system.

[0023] Example 1 See Figure 1 This embodiment provides a multi-equipment collaborative energy efficiency optimization method for prefabricated substations, the method comprising the following steps: Step S1: Collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak time periods, and generate pressure distribution data based on the energy storage level data in each peak time period.

[0024] In some embodiments, the substation distributed controller periodically acquires real-time load data and energy storage level data from data sources such as current transformers on the low-voltage side of the transformer and battery management systems of energy storage units via industrial Ethernet. The real-time load data includes the active power on the low-voltage side of the transformer, the instantaneous current of each feeder, and load statistics fed back by the distribution automation terminal; the energy storage level data includes the remaining capacity, dischargeable power, and current charging / discharging status of the energy storage units. The typical sampling period for the real-time load data is five minutes, and the typical sampling period for the energy storage level data is one minute; both are stored in the distributed controller's local time-series database.

[0025] The peak prediction process comprises two sub-steps. The first sub-step involves applying a sliding window mean filter to the real-time load data to obtain a filtered load sequence. The window length can be selected as 30 minutes to eliminate prediction noise caused by single-point sampling jumps. The sliding window mean is tuned based on the fact that the typical short-term fluctuation period of feeder loads in prefabricated substations is approximately 10 to 20 minutes; setting the window length to 1.5 to 2 times this period smooths short-term random fluctuations without masking hourly load trends. The second sub-step involves extrapolating the filtered load sequence using a pre-trained time-series predictor to obtain a predicted load sequence. The time-series predictor can employ a differential autoregressive moving average model or a prediction model based on a recurrent neural network. Its training samples are accumulated historical load data from a local time-series database, and the training set length can be selected as data from the most recent four quarters, covering different seasons and different load types of operating scenarios.

[0026] The steps for building a prediction model based on a recurrent neural network (using LSTM or GRU as typical choices) are as follows: Step 1: First, normalize the historical load sequence. MinMaxScaler is commonly used to scale the data to the [0,1] interval to eliminate dimensional differences and accelerate model convergence. Then, a sliding window method is used to transform the original sequence into a supervised learning format: set the sliding window length (e.g., predict the next time step using the past 24 time steps), and generate input-label pairs one by one. The dataset is split into training, validation, and test sets in a 7:2:1 ratio, maintaining temporal continuity.

[0027] Step 2: Build a sequential model based on Keras / TensorFlow: Add an LSTM layer (or GRU layer), specify the input shape as (window length, feature dimension), and set the return_sequences parameter to control whether the hidden state is output at each time step; to suppress overfitting, a Dropout layer can be added after the LSTM layer; finally, a fully connected layer is connected to map the hidden features to the final predicted value.

[0028] Step 3: Compile the model using the Adam optimizer and mean squared error loss function, set the batch size and number of iterations, call the fit function to train, and monitor the training process using a validation set. After training, perform parameter learning based on historical load data from the most recent four quarters in the local database, as described in the patent.

[0029] After obtaining the predicted load sequence, the continuous time periods exceeding a preset peak threshold in the predicted load sequence are identified as peak periods. The preset peak threshold can be set as the mean of the filtered load sequence plus one standard deviation, where the standard deviation is the sample standard deviation of the filtered load sequence over the past thirty days. By using the mean plus one standard deviation as the tuning method, approximately the first 16% of the high-value segments in the prediction can be identified as peak periods, which can both cover obvious high-load segments and avoid misjudging periods below the average level.

[0030] The pressure distribution data is a quantitative description of the degree of energy storage matching during each peak period, including the predicted peak load for each peak period, the available capacity of the energy storage unit in the corresponding period, and the ratio between the two, which is used for step S2 to evaluate the difference.

[0031] In another embodiment, when the computing power of the substation distributed controller is limited, the time series predictor can be replaced by a lightweight exponentially weighted moving average predictor. The calculation method is as follows: the predicted value at the current moment is obtained by adding the predicted value at the previous moment and the latest observation value according to weights, with the weight coefficient typically set to 0.3 to 0.5. The tuning basis for the weight of 0.3 to 0.5 is: too small a weight will cause the prediction to rely excessively on old observations, resulting in a slow response to load changes; too large a weight will make the prediction overly sensitive to single disturbances, leading to significant noise in the prediction sequence. Under this lightweight implementation, the distributed controller does not need to store a large number of historical load samples, but only needs to retain the predicted value at the previous moment and the observation values ​​within the window length, reducing storage requirements.

[0032] Step S2: Based on the pressure distribution data, evaluate the difference between the energy storage capacity and the demand during each peak period. If the difference exceeds a preset capacity threshold, generate an initial resource scheduling scheme according to the peak period.

[0033] The substation distributed controller compares the predicted peak load for each peak period in the pressure distribution data with the corresponding available energy storage capacity to obtain the difference. The core calculation method is as follows: First, for each peak period, the predicted peak load and the real-time available capacity of the energy storage devices are extracted from the pressure distribution data. Next, the demand gap for that period, i.e., the difference, is calculated. The distributed controller iterates through all peak periods and calculates the corresponding differences. When the difference for any period exceeds a preset capacity threshold, it is determined that the current energy storage level cannot autonomously absorb the peak load. This triggers the time-period generation procedure for the initial resource scheduling scheme, thereby quantitatively assessing the precise energy gap that the system needs to import from external sources or perform load migration.

[0034] When the difference during a peak period exceeds a preset capacity threshold, that peak period is considered to have a resource shortage. The preset capacity threshold is determined by using 10% of the rated capacity of the energy storage unit as a benchmark. The 10% value is based on the commonly used energy storage scheduling safety margin in the field. Deviations below this percentage can be absorbed by the energy storage unit's own adjustment margin without triggering a scheduling scheme reconfiguration; deviations exceeding this percentage require resource coordination on a larger scale.

[0035] Peak periods with resource shortages are sorted by the difference from largest to smallest to obtain a time period priority sequence. The substation distributed controller then performs three types of resource replenishment sequentially according to the time period priority sequence: The first category is local energy storage discharge supplementation, which allocates the dischargeable power of the energy storage unit to the current peak period; The second category is adjacent time period load migration, which shifts the load that can be postponed in the current time period (such as non-emergency charging load in electric vehicle charging piles) to the adjacent off-peak time period; The third type is external power supply, which is connected to the backup power supply of the adjacent feeder through the feeder circuit breaker.

[0036] The three types of resource replenishment are activated sequentially as described above until the difference in the current peak period is covered. The basis for this inside-out resource replenishment order is: local energy storage discharge does not rely on the neighboring network and has the fastest response; load migration is the next fastest; external power supply switching relies on circuit breaker connection operations, which takes the longest, so it is placed last.

[0037] "The difference being covered" means that the amount of supplementary resources provided by means of energy storage device discharge, load migration in adjacent periods, or external power supply is equal to or exceeds the shortfall (i.e. gap) between the energy storage capacity assessed in S2 and the peak period demand, so that the peak demand for that period can be met.

[0038] After the above resource replenishment calculations, an initial resource scheduling scheme is obtained, which includes the resource contribution ratio of each peak period and each device. The initial resource scheduling scheme includes a list of devices and their resource allocation ratios corresponding to each peak period, which are recorded as inputs for subsequent steps.

[0039] Step S3: Input the initial resource scheduling scheme into the device response simulation model to obtain the response delay sequence of each device during the peak period, and record the devices whose response delay exceeds the preset delay threshold in the operation bottleneck list.

[0040] The equipment response simulation model is pre-deployed in the substation's distributed controller. This model is constructed using an architecture of "modular unit modeling and topology network coupling." First, each individual piece of equipment within the substation (such as transformers, energy storage inverters, circuit breakers, and reactive power compensation devices) is treated as an independent simulation unit. For each unit, a dynamic system modeling approach is used to establish its input-output response characteristics. Core technical parameters in the model, such as the excitation and thermal equivalent time constants of the transformer, the power set-up time of the energy storage unit, and the mechanical dead zone of the switching devices, are precisely identified and calibrated in advance by reviewing the equipment's factory specifications and combining historical switching test data. This ensures that the dynamic behavior of individual devices after receiving dispatch instructions is highly consistent with the physical entity.

[0041] Secondly, based on unit modeling, and according to the physical topology of the substation, the communication link delay characteristics between devices and the power transmission characteristics of power tie lines are introduced as coupling boundary conditions. By dynamically splicing the input and output terminals of each independent unit according to the connection relationship of the actual electrical network and control network, a holistic simulation network covering the entire substation is finally formed. After being deployed on a distributed controller, this model can quickly simulate the transient response process of each device under the combined effects of electricity, magnetism, heat, and information flow in virtual space for the input initial resource scheduling scheme, thereby accurately outputting the operating status and response delay sequence of each device on a continuous time axis.

[0042] The substation distributed controller injects the resource allocation instructions from the initial resource scheduling scheme into the equipment response simulation model. The simulation step size can be selected as one second, and the total simulation duration covers the longest peak period. After the simulation, the equipment response simulation model outputs the response delay sequence of each device during the peak period. The response delay is the actual time taken for the device to adjust its power to the target value from receiving the scheduling instruction.

[0043] The preset delay threshold is set as follows: For the auxiliary dispatching process of prefabricated substations, the acceptable steady-state deviation time window in this field is generally ten seconds, so the delay threshold can be selected as ten seconds. For scenarios requiring faster frequency response, the delay threshold can be reduced to two seconds. When a device has samples in its response delay sequence that exceed the preset delay threshold, that device is recorded in the operational bottleneck list. The operational bottleneck list includes the device identifier, the peak time period where the bottleneck occurs, the number of samples exceeding the threshold, and the maximum magnitude of the exceedance.

[0044] In this step, the specific construction process of the equipment response simulation model includes: the equipment response simulation model adopts a discrete event simulation framework including the following three types of elements: the first type is equipment nodes, where a state machine is established for each substation device, including four states: standby, ramp-up, steady state, and fallback; the second type is communication links, where delay and bandwidth parameters are established for each pair of devices' communication links, with the initial value of the delay parameter obtained from actual measurements, and the bandwidth parameter taken as the nominal value of the communication link; the third type is environmental constraints, including temperature, humidity, and power grid harmonics. The transfer function parameters of the equipment nodes are pre-identified from historical switching test data. Specifically, each switching action in the historical switching test is extracted as an input-output pair, and the time constant and gain of the first-order or second-order transfer function are estimated using the least squares method.

[0045] The device response simulation model operates according to the following steps each time it is executed: the power command in the initial resource scheduling scheme is injected into the simulation framework in the form of discrete events; the discrete events are transmitted between the device nodes according to the delay parameters of the communication link; each device node switches its state and generates an output event based on the received event; the simulation framework processes the event queue in chronological order until the complete peak period is covered; finally, the response delay sequence of each device is extracted from the event history.

[0046] Step S4: Identify missing items in the collaborative interaction records of each device in the bottleneck list, construct a global dependency mapping between multiple devices based on the missing items, and obtain the device priority adjustment direction from the global dependency mapping.

[0047] When executing step S4, the substation distributed controller first extracts the collaborative interaction records of each device in the bottleneck list within the most recent scheduling cycle from the local communication log of the distributed controller. The collaborative interaction records are organized in the form of a task transmission matrix: the rows of the matrix represent the devices that initiate the task, the columns represent the devices that receive the task, and the elements are the number of task transmissions per unit time in that direction.

[0048] The specific operation for identifying missing items is as follows: For each row or column of the task transfer matrix, the proportion of zero elements to the total elements in that row or column is calculated. If this proportion exceeds a preset missing threshold, the device represented by the corresponding row or column is marked as a decision-making blank node. The preset missing threshold is set as follows: Under normal operating conditions where multiple devices communicate with each other within a substation, the proportion of non-zero elements in each row of the task transfer matrix is ​​generally not less than 70%. Therefore, the threshold for the proportion of zero elements is set to 30%. Exceeding this proportion indicates that the collaborative communication between this device and other devices is significantly insufficient, resulting in a decision-making blind spot.

[0049] The purpose of constructing a directed graph of dependencies is to quantitatively characterize the control coupling and information dependency strength among multiple devices in a substation using graph theory methods. Specifically, firstly, each physical device participating in coordinated scheduling within the substation (including regular nodes and identified decision-making blank nodes) is used as the topological vertex of the directed graph. Next, the aforementioned task transfer matrix is ​​traversed, and weighted directed edges are established based on the interaction directions between devices corresponding to the non-zero elements in the matrix. The edges point from the task initiating device to the task receiving device, and the edge weights are directly mapped to the number of task transfers or communication frequencies between the two devices per unit time. This constructs a directed graph of dependencies that can fully reproduce the entire substation's control coordination ecosystem.

[0050] Subsequently, the distributed controller performs a deep structural analysis of the directed graph. By performing topological sorting, it can identify and eliminate cyclic dependencies within the system that could lead to control deadlock. Simultaneously, it performs in-degree statistics on each node in the graph, precisely calculating the number of weighted directed edges pointing to that node. Based on this, and combining the sum of the weights of each incoming edge (i.e., the total frequency of received tasks), the in-degree index of each node is globally normalized, ultimately calculating the dependency score for each device. This score ranges from 0 to 1; a higher value indicates that the device relies more heavily on the cooperation of other devices during operation. The resulting global dependency mapping provides a quantitative topological basis for subsequently adjusting device priorities and addressing operational bottlenecks.

[0051] The weights refer to the value of each element in the task transfer matrix, representing the "strength" or "dependency" of task transfer from one device to another. Specifically, for a directed edge from device A to device B, the frequency, data volume, or importance of tasks or instructions transferred from A to B is quantified. The weights are obtained as follows: Based on the aforementioned "collaborative interaction records," during the actual operation of prefabricated substations, various devices (such as transformers, energy storage, and feeder protection) exchange control commands, status feedback, and load migration requests via communication networks. These interaction records are stored in a log system in time-series format.

[0052] The interactions between devices over a period of time (e.g., a complete set of peak time periods) are statistically analyzed along the "source device → target device" direction. The rows of the matrix represent source devices, the columns represent target devices, and the value of each cell is M. i,j This refers to the number of interactions from device i to device j or the cumulative task volume (such as the migration power value or the number of instructions).

[0053] The original statistical values ​​(such as interaction frequency) may vary greatly. Subsequently, it is necessary to calculate "dependency score = in-degree normalized value × total weight". Therefore, the interaction values ​​of each row or the whole are usually normalized to make the weight fall within a reasonable range, and the weight is calculated.

[0054] Normalization can be achieved using the maximum-minimum method or row-wise softmax, so that the sum of the weights of each source device to all target devices is 1, thus preserving the relative dependencies.

[0055] If a device has too many missing items in a row or column of the task transfer matrix (i.e., sparse interaction records with other devices), the device will be marked as a "decision blank node" and will need to be merged into the dependency graph separately. In this case, its weight can be filled by default value (such as a very small positive number) or based on prior knowledge of the device type.

[0056] The determination of equipment priority adjustment direction is based on the following: equipment with higher dependency scores plays a greater pivotal role in collaborative scheduling and should be given priority in obtaining resources in subsequent steps; at the same time, equipment marked as decision-making blank nodes should also be included in the positions requiring key coordination to avoid further expansion of collaborative blind spots. The equipment priority adjustment direction is output to step S5 in the form of a binary list of equipment-adjustment action (upgrade or downgrade).

[0057] The element values ​​of the task transfer matrix in this step are obtained by aggregating the local communication logs of the distributed controller by device pairing. The aggregation time window is a scheduling cycle, typically fifteen minutes. The fifteen-minute scheduling cycle is set based on the following: the regular scheduling instruction update interval of prefabricated substations is between minutes and hours. Fifteen minutes can reflect short-term changes while ensuring that the matrix elements have statistical significance; a window that is too short will cause excessive noise in the element values, while a window that is too long will mask short-term coordination anomalies.

[0058] The distributed controller maintains a pair of counters for each pair of participating devices, recording the number of forward task initiations and the number of reverse task initiations, respectively. At the end of each scheduling cycle, the distributed controller writes the counter values ​​to the corresponding positions in the task delivery matrix and then resets the counters to zero. This process can be implemented using a hash table, where the hash key is the identifier of the device pair, and the value is the forward / reverse counter.

[0059] Step S5: Based on the device priority adjustment direction, dynamically adjust the resource allocation ratio in the initial resource scheduling scheme to obtain the resource allocation coordination path.

[0060] The substation distributed controller implements dynamic weight adjustment as follows: On the one hand, the bottleneck weight of each device in the bottleneck list is normalized based on its maximum exceedance threshold. The normalization method is to divide the magnitude by the maximum exceedance threshold in the bottleneck list, with a value ranging from 0 to 1. On the other hand, the dependency score and bottleneck weight of each device are weighted and summed at a ratio of 0.6 to 0.4 to obtain a comprehensive priority score. The 0.6 to 0.4 ratio is based on the following: dependency reflects the structural position of the device in global coordination and is a longer-term indicator; bottleneck weight reflects the local urgency in the current period and is a shorter-term indicator; priority adjustment should focus on structural indicators, but at the same time retain the ability to respond to local emergencies.

[0061] The resource allocation ratios of each device in the initial resource scheduling scheme are redistributed according to the overall priority score from high to low, so that the sum of the resource allocation ratios of all devices after redistribution equals 1. The redistribution step size is 5% to 10% of the initial ratio each time, that is, the device with the higher overall priority score increases its resource allocation ratio by 5% to 10% in each round, and the corresponding resources are ceded to devices with lower overall priority. The sequence of redistributed resource allocation ratios is the resource allocation coordination path, which includes the final resource allocation ratio of each device in each peak period.

[0062] In each round, the reallocation is performed in pairs: "boosting the beneficiary device - reducing the relinquished device." The beneficiary device is the one ranked higher in the overall priority score; the relinquished device is the one ranked lower in the overall priority score and currently has the largest resource allocation ratio. The boost step in each round does not exceed 25% of the current resource allocation ratio of the relinquished device to avoid excessive relinquishment in a single instance leading to resource shortages for the relinquished device itself. The reallocation is executed cyclically until the expected load rate of the beneficiary device reaches the target level or the resource allocation ratio of all candidate relinquished devices can no longer be reduced, thereby ensuring that the sum constraint always holds in each iteration.

[0063] Step S6: Verify the resource allocation coordination path. When a local resource allocation deviation occurs, adjust the device operating parameters according to the global dependency mapping to obtain an energy efficiency optimized configuration.

[0064] The substation distributed controller injects the resource allocation coordination path into the equipment response simulation model for a second round of simulation to obtain the load rate of each device. Specifically, the resource allocation ratio during each peak period is converted into power commands for each device, injected into the simulation model, and the maximum load rate of each device during the peak period is recorded.

[0065] After the simulation, the substation distributed controller calculates the deviation between the load rate of each device and the average load rate of the device group. When the absolute value of the deviation between the load rate of at least one device and the average load rate of the device group exceeds a preset deviation threshold, it is determined that there is a local resource allocation deviation. The preset deviation threshold can be set to 10%. The 10% threshold is set based on the following: the common engineering requirement in this field for the load sharing balance of multiple devices in a prefabricated substation is that the load rate deviation of each device does not exceed 10%, and exceeding this threshold means that a significant imbalance has occurred.

[0066] When a local resource allocation deviation occurs, based on the associated device list of at least one device in the global dependency mapping, some tasks of the deviating device are migrated to the associated device. The smallest granularity of task migration is the power command within one sampling period of a single feeder. After adjustment, the simulation is repeated until the load rate deviation of all devices is within the preset deviation threshold, thereby obtaining the energy efficiency optimized configuration. The load deviation rate refers to the degree of difference between the actual load rate of the device and the preset target load rate (or the rated load capacity of the device), usually expressed as a percentage.

[0067] The energy efficiency optimization configuration includes the final operating parameters of each device in each time period, such as the target tap position of the transformer, the target charging and discharging power of the energy storage unit, and the target switching status of the feeder circuit breaker.

[0068] In this step, when the load rate deviation of at least one device exceeds a preset deviation threshold, the substation distributed controller queries the list of associated devices for that device from the global dependency mapping. The list of associated devices is sorted from high to low according to dependency scores. Task migration is attempted one by one in the list order. Each attempt migrates the power command of the smallest migration unit of the deviating device to the next associated device in the list. After migration, the simulation is run again to verify whether the load rate deviation falls within the threshold range. The power command of the smallest migration unit refers to the power scheduling amount of a single feeder in one sampling period. A maximum of three smallest units can be migrated at a time. If more than three are migrated, it is considered that the bottleneck of the device cannot be eliminated by task migration alone, and it is necessary to trigger the physical parameter adjustment of the corresponding device (e.g., change of transformer tap position or change of energy storage unit charge / discharge depth).

[0069] Step S7: Adjust the substation operation status according to the energy efficiency optimization configuration, collect operation feedback data and send it back to step S1 to update the peak prediction.

[0070] The substation distributed controller distributes the operating parameters from the energy efficiency optimization configuration to the local control units of the corresponding devices within the substation, which then adjust the parameters. During this process, the distributed controller continuously tracks the substation's operating status according to a preset acquisition cycle. The acquired operational feedback data includes the transformer's low-voltage side voltage, feeder currents, energy storage unit temperature, and the actual power consumption of each device. The preset acquisition cycle can be selected as five minutes, which is consistent with the sampling cycle of the real-time load data in step S1, facilitating the direct integration of the operational feedback data acquired in this cycle into the historical load database used in step S1.

[0071] The merged historical load database is recalled by the time-series predictor in step S1 during the next scheduling cycle, enabling the peak prediction process to reflect the latest load distribution characteristics. This closed-loop feedback mechanism gives the entire energy efficiency optimization method adaptive capabilities, and its load balancing stability is superior to that of a one-time tuned open-loop optimization scheme under long-term operation.

[0072] Example 2 This embodiment provides a multi-equipment collaborative energy efficiency optimization system for prefabricated substations. The structure of the system is as follows: Figure 2 As shown, it includes a load prediction module, a schedule generation module, a simulation testing module, a dependency building module, a weight adjustment module, a configuration determination module, and a feedback update module.

[0073] The load prediction module is used to implement step S1 in Embodiment 1, namely, to collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak time periods, and generate pressure distribution data based on the energy storage level data in each peak time period.

[0074] The scheduling generation module is used to implement step S2 in Embodiment 1, that is, to evaluate the difference between the energy storage capacity and the demand for each peak period based on the pressure distribution data, and to generate an initial resource scheduling scheme according to the peak period when the difference exceeds a preset capacity threshold.

[0075] The simulation test module is used to implement step S3 in embodiment one, that is, inputting the initial resource scheduling scheme into the device response simulation model, obtaining the response delay sequence of each device, and recording the devices whose response delay exceeds the preset delay threshold in the operation bottleneck list.

[0076] The dependency building module is used to implement step S4 in Embodiment 1, that is, to identify missing items in the collaborative interaction records of each device in the bottleneck list, build a global dependency mapping and output the device priority adjustment direction.

[0077] The weight adjustment module is used to implement step S5 in Embodiment 1, that is, to dynamically adjust the resource allocation ratio in the initial resource scheduling scheme according to the device priority adjustment direction, so as to obtain the resource allocation coordination path.

[0078] The configuration determination module is used to implement step S6 in Embodiment 1, that is, to verify the resource allocation coordination path, and adjust the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs, so as to obtain an energy efficiency optimized configuration.

[0079] The feedback update module is used to implement step S7 in Embodiment 1, that is, to adjust the substation operation status according to the energy efficiency optimization configuration and collect operation feedback data to send back to the load prediction module.

[0080] The modules described above are implemented in software and deployed in the processor of the substation distributed controller; alternatively, they can be implemented using dedicated hardware circuits, such as a field-programmable gate array (FPGA) to carry the equipment response simulation model to improve simulation speed. Data is exchanged between modules through the inter-process communication mechanism within the distributed controller.

[0081] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A multi-device collaborative energy efficiency optimization method for a prepackaged electrical substation, characterized in that, The method includes: S1. Collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak time periods, and generate pressure distribution data based on the energy storage level data in each peak time period. S2, based on the pressure distribution data, evaluate the difference between the energy storage capacity and the demand during each peak period. If the difference exceeds the preset capacity threshold, generate an initial resource scheduling scheme according to the peak period. S3, input the initial resource scheduling scheme into the device response simulation model to obtain the response delay sequence of each device during the peak period, and record the devices whose response delay exceeds the preset delay threshold in the operation bottleneck list; S4, identify missing items in the collaborative interaction records of each device in the bottleneck list, construct a global dependency mapping between multiple devices based on the missing items, and obtain the device priority adjustment direction from the global dependency mapping; S5. Based on the device priority adjustment direction, dynamically adjust the resource allocation ratio in the initial resource scheduling scheme to obtain the resource allocation coordination path; S6, verify the resource allocation coordination path, and adjust the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs, to obtain an energy efficiency optimized configuration; S7, adjust the substation operation status according to the energy efficiency optimization configuration, collect operation feedback data and send it back to step S1 to update the peak prediction.

2. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 1, characterized in that, The real-time load data is used to predict peak values, resulting in multiple peak time periods, including: The real-time load data is filtered by sliding window mean filtering to obtain the filtered load sequence. The time series predictor trained based on historical load data is extrapolated to the filtered load sequence to obtain the predicted load sequence. The continuous time period in the predicted load sequence that exceeds the preset peak threshold is determined as the peak period.

3. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 1, characterized in that, The generation of the initial resource scheduling scheme based on peak periods includes: Based on the difference between each peak period, the peak periods are sorted from largest to smallest to obtain a time period priority sequence; Resource allocation is performed sequentially according to the time period priority sequence, starting with energy storage devices, load migration in adjacent time periods, and external power supply, until the difference is covered, thus obtaining the initial resource scheduling scheme.

4. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 1, characterized in that, The step of identifying missing items in the collaborative interaction records of each device in the bottleneck list and constructing a global dependency mapping between multiple devices based on the missing items includes: Organize the collaborative interaction records of each device in the bottleneck list into a task transfer matrix, and mark the devices corresponding to rows or columns in the task transfer matrix where the number of missing items exceeds a preset missing threshold as decision blank nodes. Using each device in the substation as a node and the elements of the task transfer matrix between devices as weighted directed edges, a dependency directed graph is constructed. The decision blank nodes are merged into the directed graph of dependencies, and topological sorting and in-degree statistics are performed on the directed graph of dependencies to obtain the global dependency mapping.

5. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 4, characterized in that, The global dependency mapping includes the dependency scores of each device.

6. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 5, characterized in that, The step of dynamically adjusting the resource allocation ratio in the initial resource scheduling scheme according to the device priority adjustment direction to obtain the resource allocation coordination path includes: Based on the dependency scores and bottleneck weights of each device, a comprehensive priority score is calculated; The resource allocation ratios of each device in the initial resource scheduling scheme are redistributed according to the comprehensive priority score from high to low, and the sum of the resource allocation ratios of all devices after redistribution is equal to 1, thus obtaining the resource allocation coordination path.

7. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 1, characterized in that, The step of verifying the resource allocation coordination path and adjusting the device operating parameters based on the global dependency mapping when local resource allocation deviations occur includes: The resource allocation and coordination path is simulated a second time in the device response simulation model to obtain the load rate of each device. When the absolute value of the deviation between the load rate of at least one device and the average load rate of the device group exceeds a preset deviation threshold, it is determined that there is a local resource allocation deviation, and the device operating parameters are adjusted according to the associated devices of the at least one device in the global dependency mapping.

8. The multi-equipment collaborative energy efficiency optimization method for prefabricated substations according to claim 1, characterized in that, The process of collecting and sending operational feedback data back to step S1 to update the peak prediction includes: The voltage, current, and temperature data of the substation's operating status are collected according to a preset collection cycle as the operating feedback data; The operational feedback data is incorporated into the historical load database, so that the peak prediction prediction model is used to calculate the peak in the next scheduling cycle using the updated historical load data that includes the operational feedback data.

9. A multi-equipment collaborative energy efficiency optimization system for a prefabricated substation employing the multi-equipment collaborative energy efficiency optimization method for prefabricated substations as described in any one of claims 1 to 8, characterized in that, The system includes: The load prediction module is used to collect real-time load data and energy storage level data of multiple devices in the substation, perform peak prediction on the real-time load data, divide it into multiple peak periods, and generate pressure distribution data based on the energy storage level data in each peak period. The scheduling generation module is used to evaluate the difference between the energy storage capacity and the demand during each peak period based on the pressure distribution data, and generate an initial resource scheduling scheme according to the peak period when the difference exceeds a preset capacity threshold. The simulation test module is used to input the initial resource scheduling scheme into the device response simulation model, obtain the response delay sequence of each device, and record the devices whose response delay exceeds the preset delay threshold into the operation bottleneck list. The dependency building module is used to identify missing items in the collaborative interaction records of each device in the list of operational bottlenecks, build a global dependency mapping, and output the direction of device priority adjustment. The weight adjustment module is used to dynamically adjust the resource allocation ratio in the initial resource scheduling scheme according to the device priority adjustment direction, so as to obtain the resource allocation coordination path. The configuration determination module is used to verify the resource allocation coordination path, and adjust the device operating parameters according to the global dependency mapping when a local resource allocation deviation occurs, so as to obtain an energy efficiency optimized configuration. The feedback update module is used to adjust the substation's operating status according to the energy efficiency optimization configuration and collect operating feedback data to send back to the load prediction module.

10. The multi-equipment collaborative energy efficiency optimization system for prefabricated substations according to claim 9, characterized in that, In the feedback update module, voltage, current, and temperature data of the substation's operating status are collected according to a preset collection cycle as the operating feedback data, and the operating feedback data is incorporated into the historical load database for updating.

Images