Energy equipment cooperative control method and system based on intelligent fusion terminal

Abstract

The application discloses an energy equipment cooperative control method and system based on an intelligent fusion terminal, relates to the technical field of smart grid dispatching, and comprises the following steps: collecting operation state data of a plurality of distributed energy equipment in a target control area and rearranging the operation state data multiple times to generate a plurality of arrangement sample sequences; inputting the arrangement sample sequences into a power accumulation engine built in the intelligent fusion terminal; extracting accumulator numerical value changes and accumulator final values to obtain a plurality of decision feature vectors; selecting a plurality of representative decision feature vectors through unsupervised clustering, and constructing a response consistency preference data pair set; and deploying the power accumulation engine back to the intelligent fusion terminal after optimizing and training the power accumulation engine by using the set, so as to realize real-time cooperative control of the distributed energy equipment. The application solves the technical problem that random reporting sequences of the distributed energy equipment lead to inconsistent control results, and achieves the technical effect that the reporting sequence of the equipment no longer affects the calculation result under the same physical state, and the terminal outputs unique, stable and accurate adjustable power.

Description

Technical Field

[0001] This invention relates to the field of smart grid dispatching technology, and in particular to a method and system for collaborative control of energy equipment based on a smart fusion terminal. Background Technology

[0002] Intelligent converged terminals are widely used in the coordinated control of distributed energy resources. Their data processing and power calculation results are crucial for grid dispatch safety, stable equipment operation, and efficient energy management. The terminal's ability to uniformly process data reported from multiple devices is the core of ensuring precise control. In existing technologies, intelligent converged terminals mostly use built-in power accumulation engines for data processing and adjustable power calculation, enabling basic control in fixed-time reporting scenarios. However, in reality, device reporting times are random and disordered. Existing engine data processing logic is highly sensitive to the reporting order. Under the same operating conditions, different reporting orders result in significant deviations in the adjustable power output, leading to inconsistent and inaccurate data. These problems prevent the terminal from outputting stable and reliable control data, making it difficult to meet the actual needs of precise coordinated control of distributed energy resources and safe grid dispatch. Summary of the Invention

[0003] This application provides a collaborative control method and system for energy equipment based on an intelligent fusion terminal, which solves the technical problem of inconsistent control results caused by the random reporting order of distributed energy equipment.

[0004] The first aspect of this application provides a method for collaborative control of energy devices based on an intelligent fusion terminal. The method includes: collecting multiple operating status data of multiple distributed energy devices within a target control area, and rearranging the multiple operating status data multiple times to generate multiple permutation sample sequences; inputting the multiple permutation sample sequences sequentially into a power accumulation engine built into the intelligent fusion terminal, extracting the accumulator value changes and the final value of the accumulator after processing each permutation sample sequence, to obtain multiple decision feature vectors; performing unsupervised clustering on the multiple decision feature vectors to obtain multiple cluster sets, and selecting multiple representative decision feature vectors from each of the obtained cluster sets; constructing a response consistency preference data pair set based on the matching relationship between the multiple representative decision feature vectors and a preset collaborative control target; using the response consistency preference data pair set to perform preference optimization training on the power accumulation engine, and deploying the trained power accumulation engine back to the intelligent fusion terminal for real-time collaborative control of the multiple distributed energy devices.

[0005] A second aspect of this application provides a collaborative control system for energy devices based on an intelligent fusion terminal. The system includes: a permutation sample sequence generation module, used to collect multiple operating status data of multiple distributed energy devices within a target control area, and to rearrange the multiple operating status data multiple times to generate multiple permutation sample sequences; a decision feature vector acquisition module, used to input the multiple permutation sample sequences sequentially into a power accumulation engine built into the intelligent fusion terminal, extracting the accumulator value changes and the final value of the accumulator after processing each permutation sample sequence to obtain multiple decision feature vectors; a representative vector selection module, used to traverse the multiple decision feature vectors to perform unsupervised clustering, obtain multiple cluster sets, and select multiple representative decision feature vectors from the obtained cluster sets; a preference data pair set construction module, used to construct a response consistency preference data pair set based on the matching relationship between the multiple representative decision feature vectors and a preset collaborative control target; and a real-time collaborative control execution module, using the response consistency preference data pair set to perform preference optimization training on the power accumulation engine, deploying the trained power accumulation engine back to the intelligent fusion terminal, and performing real-time collaborative control of the multiple distributed energy devices.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application collects operational status data from distributed energy devices and, through data processing steps such as sequence rearrangement, feature extraction, unsupervised clustering, and preference optimization, obtains engine decision features and response consistency preference data. It then constructs preference data pairs and trains a power accumulation engine, thereby eliminating the influence of the device data reporting order on the calculation results. This enables the intelligent fusion terminal to output a stable and unified power value, making the collaborative control of energy devices more precise and reliable. Under the same physical state, the device reporting order no longer affects the calculation results, and the terminal outputs a unique, stable, and accurate adjustable power. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the collaborative control method for energy equipment based on an intelligent fusion terminal provided in an embodiment of this application.

[0010] Figure 2This is a schematic diagram of the structure of the energy equipment collaborative control system based on the intelligent fusion terminal provided in the embodiments of this application.

[0011] Figure labeling: Module 1 for generating sample sequences, Module 2 for obtaining decision feature vectors, Module 3 for selecting representative vectors, Module 4 for constructing preference data pair sets, and Module 5 for executing real-time collaborative control. Detailed Implementation

[0012] This application provides a collaborative control method and system for energy equipment based on an intelligent fusion terminal, which solves the technical problem of inconsistent control results caused by the random reporting order of distributed energy equipment.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, a collaborative control method for energy equipment based on an intelligent fusion terminal is provided, wherein the method includes: Multiple operational status data of multiple distributed energy devices within the target control area are collected, and the multiple operational status data are rearranged multiple times to generate multiple permutation sample sequences.

[0016] In this embodiment of the application, the target control area is the power distribution site where various types of distributed energy equipment such as distributed photovoltaic inverters, energy storage converters, charging piles and interruptible load circuit breakers are deployed. Typical scenarios include industrial parks, commercial buildings or residential community transformer areas.

[0017] Specifically, firstly, within the target control area, intelligent fusion terminals serve as the core data acquisition devices. Based on the actual installation locations of distributed photovoltaic inverters, energy storage converters, charging piles, and interruptible load circuit breakers, the intelligent fusion terminals are deployed in the physical center areas of each device. The wired transmission distance between the intelligent fusion terminal and a single distributed energy device does not exceed 1000 meters. The deployment environment avoids strong electromagnetic interference sources. Simultaneously, a bidirectional communication link is established with the upper-level power grid dispatch automation system via industrial Ethernet. An RS485 industrial communication bus is used to connect the communication interfaces of the intelligent fusion terminals and various distributed energy devices in series. A stable data transmission channel is established using the Modbus RTU communication protocol. Each distributed energy device is assigned a unique 1-byte device address, with an address value ranging from 1 to 247. The communication frame format uses 1 start bit, 8 data bits, 1 stop bit, and 1 even parity bit to ensure stable communication and accurate data verification between the devices and the terminal.

[0018] Then, the intelligent fusion terminal broadcasts a collection command to all distributed energy devices on the bus according to a fixed collection cycle, such as 100ms. After receiving the collection command, each distributed energy device collects and generates operating status data of various types of distributed energy devices in real time through its built-in metering sensing unit, operation monitoring unit, and control unit. This operating status data is a basic parameter that all devices have physical collection capabilities for and is necessary for subsequent power calculation and coordinated control. Specifically, it includes device type code, real-time active power, maximum adjustable power, minimum adjustable power, and response time in milliseconds.

[0019] The equipment type code is a preset 2-digit decimal fixed code for each device. For example, distributed photovoltaic inverters correspond to code 01, energy storage converters to code 02, charging piles to code 03, and interruptible load circuit breakers to code 04. Real-time active power is obtained by the metering sensor unit built into each device, which samples and acquires the data in real time according to the metering rules of the power industry standard DL / T 645. The maximum adjustable power and minimum adjustable power are calculated and generated in real time by the control unit of each device based on the current operating conditions, hardware rated parameters, and protection thresholds. The response time in milliseconds is generated by the built-in timing unit of each device, which calculates the time interval from the moment the acquisition command is received to the moment the parameter acquisition is completed. For energy storage converters with additional parameter acquisition capabilities, the state of charge percentage is collected synchronously as a supplementary parameter and generated along with the common operating status data. Other devices without corresponding acquisition capabilities do not need to generate this supplementary parameter. The operating status data generated by all devices follows a fixed field order and data length to ensure that the data format is completely uniform.

[0020] Next, after each distributed energy device generates its operating status data, it is asynchronously uploaded to the intelligent fusion terminal via an RS485 communication bus. Upon receiving a single piece of operating status data, the intelligent fusion terminal marks it with a millisecond-precision timestamp and stores the data in its local buffer. The intelligent fusion terminal preprocesses the received data in the buffer. For duplicate reports from the same device address and within the same timestamp, only the most recently received data is retained. Invalid data with missing common operating status fields or values ​​exceeding the device's rated range is directly filtered and discarded; missing supplementary parameters do not affect data validity. Data that times out beyond twice the acquisition period is discarded. After preprocessing, the valid operating status data is stored in the terminal's local storage unit, completing the overall acquisition of common operating status data from multiple distributed energy devices within the target control area.

[0021] Finally, when generating permutation sample sequences by repeatedly rearranging multiple running status data, it is determined whether the number of data is less than or equal to the first quantity threshold. If so, its factorial permutations are enumerated as samples. Otherwise, random permutations of the preset first value are generated by random permutation, and then permutations of the preset second value are selected according to the principle of maximizing the number of element position exchanges between permutations. Finally, multiple permutation sample sequences are obtained. The specific implementation process of this step will be described in detail below.

[0022] Through standardized terminal deployment, communication configuration, parameter acquisition, data generation and preprocessing, the entire process design provides stable, unified and effective data support for subsequent data processing and equipment collaborative control.

[0023] Multiple permutation sample sequences are sequentially input into the power accumulation engine built into the intelligent fusion terminal. The changes in the accumulator values ​​and the final value of the accumulator after processing each permutation sample sequence are extracted to obtain multiple decision feature vectors.

[0024] In this embodiment, the intelligent fusion terminal is an edge computing device installed in scenarios such as power distribution substations. It integrates functions such as power supply and consumption information collection, equipment status monitoring, communication networking, local analysis and decision-making, and collaborative control, supporting the interaction between power generation, grid, load, and storage, as well as the intelligent collaborative management of energy equipment.

[0025] Optionally, after generating and caching all permutation sample sequences, the intelligent fusion terminal stores all permutation sample sequences in the cache into a built-in first-in-first-out (FIFO) processing queue according to their generation order. This queue adopts a linear queue management method commonly used in industrial embedded systems, ensuring that the processing order of the permutation sample sequences is completely consistent with their generation order. Then, the intelligent fusion terminal sequentially retrieves a single permutation sample sequence from the head of this processing queue. Simultaneously, it initializes all four accumulators in the power accumulation engine to 0. Then, according to the inherent arrangement order of the running status data within the permutation sample sequence, it inputs each running status data point into the power accumulation engine to perform power accumulation processing.

[0026] Next, after the power accumulation engine processes each single piece of running status data, it records the current values ​​of the four accumulators to form a stage vector and puts it into the corresponding stage vector table. After a single permutation sample sequence is processed, the stage vector table and the final value of the accumulator are concatenated to generate the corresponding decision feature vector and summarize to obtain multiple decision feature vectors. This step will be explained in detail in the following content.

[0027] Unsupervised clustering is performed by traversing multiple decision feature vectors to obtain multiple cluster sets, and multiple representative decision feature vectors are selected from the obtained multiple cluster sets respectively.

[0028] In one embodiment of this application, a distance matrix is ​​first constructed by calculating the Euclidean distance between each decision feature vector. Each vector is then used as an initial independent cluster. The clusters corresponding to the smallest Euclidean distance in the distance matrix are repeatedly merged and the distance matrix is ​​updated until the largest Euclidean distance within a cluster exceeds a preset threshold. This completes unsupervised agglomerative hierarchical clustering and obtains multiple cluster sets. The specific implementation process of this step will be described in detail below.

[0029] Finally, for each cluster set, iterate through all the decision feature vectors in it, calculate the average Euclidean distance between each vector and all other vectors in the cluster, and select the decision feature vector with the smallest average distance as the representative decision feature vector of the cluster set, thus obtaining multiple representative decision feature vectors.

[0030] Based on the matching relationship between the multiple representative decision feature vectors and the preset collaborative control target, a set of response consistency preference data pairs is constructed.

[0031] Specifically, the target power value issued by the power grid dispatch is used as the preset collaborative control target. The final value of the accumulator of each representative decision feature vector is compared and the sum of the four-dimensional power absolute deviation is calculated. Based on the magnitude of the deviation, positive and negative preference clusters are divided. Then, corresponding vector combinations are selected to construct response consistency preference data pairs. This step will also be explained in detail in the following content.

[0032] Using the set of response consistency preference data pairs, preference optimization training is performed on the power accumulation engine. The trained power accumulation engine is then deployed back to the intelligent fusion terminal to perform real-time collaborative control of the multiple distributed energy devices.

[0033] Specifically, firstly, using the response consistency preference data pair set as the supervision basis, the initial values ​​of the four accumulators of the power accumulation engine and the power accumulation coefficients of various devices are set as trainable parameters. The parameters are iteratively updated through the pairwise ranking loss function to achieve the corresponding optimization objective. Training stops when the standard deviation of the final value of the accumulator in the positive preference cluster set meets the limiting condition, and the parameters are saved to obtain the trained power accumulation engine. The specific implementation process of this step will be described in detail below.

[0034] Next, the trained power accumulation engine parameters, stored in a fixed format, are written into the real-time computing program of the intelligent fusion terminal, completing the terminal deployment of the engine and replacing the original power accumulation engine, thus entering real-time collaborative control operation. During real-time operation, the intelligent fusion terminal continuously acquires real-time operating status data from multiple distributed energy devices within the distribution area through its acquisition module. There is no need to sort, organize, prioritize, or correct the timing of the collected operating status data; regardless of the order in which the device data is reported, it is directly and indiscriminately input into the trained power accumulation engine in the original order of acquisition.

[0035] The power accumulation engine, based on fixed training parameters, performs real-time accumulation calculations on the input operating status data. After each data point is matched and accumulated, the real-time values ​​of the four accumulators are updated synchronously. Once all device data within the current acquisition cycle has been input, the final values ​​of the four accumulators are directly output. These final values ​​serve as the basis for real-time power regulation in the coordinated control of the power grid, distribution network, load, and energy storage system within the distribution area. The intelligent fusion terminal, based on the output accumulator final values, performs difference calculations in the same dimension with the preset coordinated control target to obtain the regulation deviation in each power dimension. Then, according to the real-time adjustable capacity ratio of similar devices within the distribution area, the regulation deviation is allocated to each distributed energy device, generating corresponding power regulation commands. These commands adjust the upper limit of output for photovoltaic inverters, adjust the charging and discharging power setpoints for energy storage converters, and adjust the switching status and power limits for charging piles and interruptible load circuit breakers. Finally, the regulation commands are sent to the corresponding distributed energy devices through a dedicated power communication channel, completing the real-time coordinated control of multiple distributed energy devices.

[0036] Furthermore, the method provided in this application embodiment includes: If the number of multiple running state data is less than or equal to a first quantity threshold, then enumerate the factorial permutations of the multiple running state data and take each permutation as a permutation sample sequence to obtain multiple permutation sample sequences; otherwise, use a random permutation algorithm to generate a random permutation of a preset first value, and then select a permutation of a preset second value according to the principle of maximizing the number of element position swaps between permutations to obtain multiple permutation sample sequences, wherein the preset first value is greater than the preset second value.

[0037] Optionally, the first quantity threshold is pre-set based on the correspondence between the computing power of the intelligent fusion terminal and the computational volume of permutations. Specifically, the acquisition process involves first using the processor performance testing program built into the intelligent fusion terminal to test the total number of permutation samples that the terminal can process in a single cycle. The duration of this single cycle is consistent with the data acquisition cycle of the running status data. Then, based on the characteristic that the number of permutations increases factorially with the number of data points, the number of permutations corresponding to different data points is calculated. The maximum number of data points whose permutation count does not exceed the total number of data points that the terminal can process in a single cycle is selected as the first quantity threshold. For example, the intelligent fusion terminal uses an industrial-grade general-purpose MCU, which can process 50,000 permutation samples within a single acquisition cycle of 100ms. Calculations show that the number of permutations corresponding to 8 running status data points is 40,320, and the number of permutations corresponding to 9 running status data points is 362,880. Therefore, 8 is selected as the first quantity threshold. This threshold can be adaptively adjusted according to the computing power of the terminal processor.

[0038] Simultaneously, preset first and second values ​​are established. Both values ​​are based on the total number of permutation samples that the terminal can process in a single cycle, with the first preset value always being greater than the second preset value. The second preset value is the target number of permutation sample sequences that the terminal can process in a single cycle. Its core setting rule is that the upper limit of the value does not exceed the total number of permutation samples that the terminal can process in a single cycle, and it is on the same order of magnitude as the number of permutations corresponding to the first threshold. That is, the difference between the two values ​​is controlled within one order of magnitude, ensuring that the sample input scale processed by the subsequent algorithm is consistent and the terminal's computing load is stable under different data volumes. For example, the terminal can process a total of 50,000 groups per cycle, and the number of permutations corresponding to the first quantity threshold of 8 is 40,320 groups. The preset second value is 40,000 groups. The preset first value is a multiplier derived from the preset second value. The multiplier is selected in the range of 5-20 times, preferably 10 times. This multiplier range can take into account both the sufficiency of sample screening and the terminal's computing load. For example, the preset first value is 400,000 groups, corresponding to the preset second value of 40,000 groups. Both values ​​can be adjusted synchronously according to the terminal's computing power and sample coverage requirements.

[0039] Next, after the intelligent fusion terminal completes the collection and preprocessing of the operating status data, the total number of valid operating status data is counted. This total number corresponds to the total number of distributed energy devices that have actually completed valid reporting. Then, the total number of devices is compared with the first quantity threshold for judgment. The specific steps are as follows: If the total number of running status data is less than or equal to the first quantity threshold, the lexicographical permutation algorithm is used to enumerate all factorial permutations of the running status data. Specifically, the process involves first arranging all running status data in ascending order of their received timestamps to generate an initial sequence. Then, the lexicographical permutation algorithm is used to sequentially generate all unique permutations. Each generated permutation is stored as an independent permutation sample sequence in the terminal buffer until all factorial permutations are generated, resulting in a complete set of permutation sample sequences.

[0040] If the total number of running status data exceeds a first threshold, all running status data are first sorted in ascending order by receiving timestamp to generate an initial basic sequence. Then, the Fisher-Yates random permutation algorithm is used to generate a random permutation of a preset first value. During algorithm execution, the random number generation uses the hardware random number generator built into the industrial-grade MCU of the intelligent fusion terminal. The random number register is read through the MCU standard library function to obtain an integer random number between 0 and the target range, ensuring that the generated random numbers are evenly distributed and without periodic deviation. The specific execution process of the algorithm is as follows: First, obtain the total number of elements in the initial basic sequence. Starting from the last element of the sequence, traverse each element in the sequence backward. For the i-th element currently traversed, generate an integer random number between 0 and i-1 using the random number generator. Using this random number as an index, select the element with the corresponding index from the first i elements of the sequence. Swap the position of the currently traversed element with the randomly selected element. After completing one complete traversal, a random permutation is generated. This algorithm process is repeated until a random permutation of the preset first value is generated. All generated random permutations are stored in the terminal's temporary buffer.

[0041] After generating the random permutations, permutations with a preset second value are selected from the random permutations in the temporary buffer according to the principle of maximizing the number of element position swaps between permutations. The specific execution process is as follows: First, a permutation is randomly selected from the temporary buffer as the initial selected permutation and stored in the selection result set. Then, all remaining random permutations in the temporary buffer are traversed. For each permutation to be selected, the total number of element position swaps between it and all selected permutations in the selection result set is calculated. The total number of element position swaps is calculated by counting the total number of identical elements in different positions between the permutation to be selected and a single selected permutation. This total number is the number of element position swaps between the two permutations. The total number of swaps for the permutation to be selected is then added to the total number of swaps for that permutation. After each full traversal, the permutation with the largest total number of swaps is selected and added to the selection result set. This traversal and selection process is repeated until the number of permutations in the selection result set reaches the preset second value. Each permutation in the selection result set is then treated as a permutation sample sequence, ultimately resulting in multiple permutation sample sequences.

[0042] By clarifying the threshold determination rules, numerical setting basis, random number generation mechanism, random permutation algorithm execution steps, and differential sample screening logic, the entire process of generating permutation sample sequences is fully explained. This ensures complete scenario coverage of all permutations under small data volume and achieves maximum differential coverage of sample sequences under large data volume, providing comprehensive and effective sample support for eliminating the impact of data reporting order on power calculation results.

[0043] Furthermore, the method provided in this application embodiment includes: The power accumulation engine includes four accumulators: a photovoltaic power accumulator, an energy storage discharge power accumulator, an energy storage charging power accumulator, and an interruptible load power accumulator.

[0044] Specifically, the power accumulation engine is a software processing module running within the intelligent fusion terminal. It uses four independent accumulation variables to classify and accumulate power data: photovoltaic power accumulator, energy storage discharge power accumulator, energy storage charging power accumulator, and interruptible load power accumulator. Each accumulation variable is initialized to 0 before processing a new permutation sample sequence to ensure that the accumulation calculations of different permutation sample sequences are independent of each other.

[0045] There is a clear correspondence between the accumulator and the operating status data of the distributed energy equipment. The power accumulation engine has a pre-set power accumulation coefficient for each type of equipment. The initial default value of the coefficient is 1. The power value of the operating status data needs to be multiplied by the power accumulation coefficient of the corresponding equipment type to complete the weighting process, and then input into the specified accumulator according to the corresponding rules. The accumulation input rules for the power accumulation engine are as follows: Data from distributed photovoltaic inverters with device type code 01 is input only to the photovoltaic power accumulator after weighted processing; data from energy storage converters with device type code 02 is input based on the positive or negative status of real-time active power. When the real-time active power is greater than 0, it indicates that the energy storage is in a discharging state, and the power value is input to the energy storage discharge power accumulator after weighted processing; when the real-time active power is less than 0, it indicates that the energy storage is in a charging state, and the absolute value of the power value is input to the energy storage charging power accumulator after weighted processing; when the real-time active power is equal to 0, it indicates that the energy storage is in a standby state, and the data is not input to any accumulator and is skipped directly; data from charging piles with device type code 03 and data from interruptible load circuit breakers with device type code 04 are both input to the interruptible load power accumulator after weighted processing.

[0046] When the power accumulation engine processes the permutation sample sequence, it reads each data point sequentially according to the order of the operating status data in the sequence. Based on the equipment type code and power status in the data, it inputs the real-time active power value into the corresponding accumulator variable. After completing the reading and input of all data in the current permutation sample sequence, the current values ​​of the photovoltaic power accumulator, energy storage discharge power accumulator, energy storage charging power accumulator, and interruptible load power accumulator are the classification accumulation power results under this permutation order. This result will be used as the input data for the subsequent feature extraction steps.

[0047] By employing a categorized accumulation structure with four accumulators, independent statistics on the power contribution of different types of equipment are achieved. This ensures that each piece of operational status data is accurately assigned to its corresponding power dimension, providing a stable and structured input foundation for subsequent analysis of the impact of different data arrangement orders on the power accumulation results. Furthermore, this comprehensively covers the power accumulation requirements of all distributed energy devices within the target control area, supporting the subsequent optimization process to eliminate the influence of data reporting order on the coordinated control results.

[0048] Furthermore, the method provided in this application embodiment includes: After the power accumulation engine processes one running status data in each permutation sample sequence, it records the current values ​​of the four accumulators as a stage vector in the stage vector table of that permutation sample sequence. When the permutation sample sequence is finished, it concatenates the stage vector table of the permutation sample sequence with the final value of the accumulators of the permutation sample sequence to obtain the decision feature vector of the permutation sample sequence, and adds it to multiple decision feature vectors.

[0049] Specifically, after each data point is matched and processed by the power accumulation engine, a value reading operation is immediately triggered. Regardless of whether the data has undergone accumulation calculation, the current values ​​of the stored variables corresponding to the four accumulators are obtained through the variable reading instructions common to industrial-grade MCUs. The real-time values ​​of each accumulator are the cumulative results after weighted calculation using the corresponding device's power accumulation coefficient. If the data conforms to the accumulation input rules in the aforementioned steps, the value is read after allocation matching and accumulation calculation. If the data is zero-power data in the standby state of the energy storage converter, the accumulation process is skipped according to the rules, and the currently unchanged accumulator value is read directly. After reading, the four values ​​are combined into a one-dimensional stage vector in a fixed order: photovoltaic power accumulator, energy storage discharge power accumulator, energy storage charging power accumulator, and interruptible load power accumulator. The number of elements in each stage vector is fixed at four. The intelligent fusion terminal pre-allocates a contiguous storage space as a stage vector table for the currently processed permutation sample sequence. The generated stage vectors are stored in the stage vector table in the order of data processing. The total number of stage vectors stored in the stage vector table is completely consistent with the total number of running status data entries contained in the current permutation sample sequence, with no omissions or redundancies.

[0050] Next, after processing all operational status data within the current permutation sample sequence, the intelligent fusion terminal reads the final values ​​of the four accumulators again. Following a fixed order identical to the stage vectors, these values ​​are combined into a one-dimensional accumulator final value vector, with a fixed number of four elements. Then, using a one-dimensional array concatenation method, all stage vectors stored in the stage vector table are first expanded sequentially in the order of insertion and concatenated end-to-end to form a complete long one-dimensional vector. The accumulator final value vector is then appended to the end of this long one-dimensional vector, forming a decision feature vector uniquely corresponding to the current permutation sample sequence. This generation rule ensures that for the same batch of validly reported device data, regardless of the permutation order, all generated decision feature vectors have completely uniform dimensions with no dimensional differences. After generating each decision feature vector, the intelligent fusion terminal stores it in its local feature vector set. Once all permutation sample sequences in the processing queue have been processed, multiple decision feature vectors corresponding to the number of permutation sample sequences are obtained.

[0051] By using a standardized feature generation method that records the entire process of stage vectors corresponding to all data entries and concatenates the final values, the dynamic impact of different data arrangement orders on the entire power accumulation process and the differences in the final results are fully captured. This provides standardized feature input with unified dimensions and no omissions for subsequent engine optimization, effectively avoiding the problem of missing process features caused by using only the accumulator's final value.

[0052] Furthermore, the method provided in this application embodiment includes: The number of rows in the stage vector table is equal to the number of records of running state data in the permuted sample sequence, and each row contains four accumulator values.

[0053] Specifically, the intelligent fusion terminal relies on the structured data storage method built into the embedded system, using a two-dimensional array as the underlying storage carrier to construct a stage vector table. For each permutation sample sequence to be processed, a separate blank data table is pre-created to hold the stage vector data. Then, the terminal counts the total number of valid running status data in the current permutation sample sequence, and uses this number as the basis for the total row layout of the data table. The total number of rows in the data table is strictly consistent with the number of running status data entries, and quantitative matching and layout are completed based on the row structure of the two-dimensional array.

[0054] Next, we will uniformly set the data storage specifications for each row of the data form. Each row storage unit will hold four numerical values, which will correspond to the real-time calculation values ​​of four independent accumulators. We will maintain the uniformity of the data dimension of a single row by leveraging the fixed number of columns of a two-dimensional array.

[0055] The power accumulation engine arranges the operating status data within the sample sequence one by one, performs weighted calculations and corresponding accumulation processing according to the power accumulation coefficients corresponding to each device type, and immediately reads the real-time values ​​of the four accumulators after each data processing step is completed. The data is then filled into a single row of the stage vector table in a preset fixed order, and the data entry is completed row by row according to the order of data processing.

[0056] By standardizing the table's row and column layout and fixing the single-row data configuration, and relying on two-dimensional arrays, the process data is stored in an orderly manner, ensuring that the stage vector record format is uniform and standardized, and providing a standardized data foundation for the subsequent stable splicing and unified operation of feature vectors.

[0057] Furthermore, the method provided in this application embodiment includes: Calculate the Euclidean distance between every two decision feature vectors in multiple decision feature vectors to construct a distance matrix; treat each decision feature vector as an independent cluster, find the two independent clusters with the smallest Euclidean distance in the distance matrix and merge them to obtain a merged cluster; delete all Euclidean distances in the distance matrix related to the two independent clusters corresponding to the merged cluster, recalculate the Euclidean distances between the merged cluster and the remaining multiple decision feature vectors excluding the merged cluster, and update the distance matrix based on the calculation results; repeat agglomerative hierarchical clustering on the updated distance matrix until the maximum Euclidean distance within each merged cluster exceeds a preset threshold, then stop clustering to obtain multiple cluster sets.

[0058] In one embodiment, the intelligent fusion terminal first iterates through all decision feature vectors. For each pair of distinct decision feature vectors, it calculates the distance between the two vectors using the standard Euclidean distance calculation method. The calculation process involves taking the difference between the corresponding dimensions of the two vectors sequentially, squaring the differences, summing them up, and then taking the square root of the sum to obtain the Euclidean distance between the two vectors. All the calculated Euclidean distances are then arranged in the index order of the decision feature vectors to construct a symmetric two-dimensional distance matrix. The number of rows and columns in the matrix is ​​equal to the total number of decision feature vectors, and the diagonal elements are set to zero, representing the distance between a vector and itself.

[0059] Next, unsupervised clustering is performed, with the following steps: Initially, each decision feature vector is treated as an independent cluster, containing only one decision feature vector. A unique identifier is assigned to each cluster, corresponding one-to-one with the index of the decision feature vector for easy identification in subsequent operations. Then, all non-zero elements in the upper triangular region of the current distance matrix are traversed to find the smallest Euclidean distance value. The two independent clusters corresponding to this value are the closest clusters. When multiple minimum Euclidean distances exist, the cluster pair with the smallest index number is selected and merged into a single group. All original decision feature vectors within this group are retained, their values ​​unchanged, only the grouping relationship is altered. The two selected independent clusters are then merged into a new cluster containing all decision feature vectors from the original two clusters, and a new unique identifier is assigned to the new cluster.

[0060] Next, all rows and columns corresponding to the two old clusters being merged are deleted from the distance matrix, eliminating distance information related to the old clusters. Then, the average link method is used to calculate the Euclidean distance between the merged new cluster and all remaining independent clusters that were not merged. That is, the distance between the new cluster and the remaining clusters is equal to the average of the Euclidean distances between the two old clusters and the remaining clusters before the merge. The newly calculated distance values ​​are then filled into the corresponding positions in the distance matrix to form the updated distance matrix.

[0061] Finally, a preset threshold corresponding to the clustering stopping condition is pre-set. This threshold is based on the dimension of the decision feature vector and the power value fluctuation range of a single operating status data point, and its value ranges from 0.5% to 2% of the maximum rated power of a single device. The threshold can be adaptively adjusted according to the terminal's computing power and the requirements of collaborative control accuracy. The operations of finding the minimum Euclidean distance, merging clusters, and updating the distance matrix are repeatedly executed. After each merge, the maximum Euclidean distance within all current clusters is checked, i.e., the maximum distance between any two decision feature vectors within a cluster. When the maximum Euclidean distance within any cluster exceeds the preset threshold, the cluster merging operation is stopped, resulting in multiple stable cluster sets.

[0062] By using an agglomerative hierarchical clustering method with clearly defined merging rules and quantified stopping thresholds, decision feature vectors under different arrangement orders are unsupervisedly grouped and representative vectors are selected. This effectively merges similar power accumulation process data, reduces the data processing scale of subsequent engine optimization, and fully preserves the key difference features under different arrangement orders.

[0063] Furthermore, the method provided in this application embodiment includes: Obtain the target power value issued by the power grid dispatch automation system and use it as the preset collaborative control target; compare the final values ​​of the accumulators in multiple representative decision feature vectors with the preset collaborative control target, and calculate the sum of absolute deviations in four power dimensions; label the cluster set corresponding to the representative decision feature vector with the smallest sum of absolute deviations as the positive preference cluster set, and label the cluster sets corresponding to the remaining representative decision feature vectors as the negative preference cluster set; select one decision feature vector from the positive preference cluster set as the winner, and select one decision feature vector from the negative preference cluster set as the loser, and construct a response consistency preference data pair set.

[0064] In this embodiment, the power grid dispatch automation system is a core power grid control platform that uses SCADA / EMS as its core and collects power grid operation data, monitors status, analyzes and makes decisions, and issues control commands in real time through a hierarchical distributed architecture to ensure the safe, stable and economical operation of the entire process of power generation, transmission and distribution.

[0065] Optionally, the intelligent fusion terminal is equipped with a dedicated power communication and interaction module. In accordance with the conventional 104 protocol data transmission protocol of the power system, it continuously receives the power regulation data of the distribution area issued by the power grid dispatch automation system, parses and extracts the target power values ​​corresponding one-to-one with the four accumulator dimensions, namely the target power of photovoltaic grid connection, the target power of energy storage discharge, the target power of energy storage charging, and the target power of interruptible load regulation. The target power values ​​of this fixed dimension are set as preset collaborative control targets and cached in the local readable storage area of ​​the terminal for later retrieval.

[0066] Then, all the selected representative decision feature vectors are retrieved sequentially. The final values ​​of the four accumulators concatenated at the end of each representative decision feature vector are independently decomposed, corresponding to the four statistical dimensions of photovoltaic power accumulator, energy storage discharge power accumulator, energy storage charging power accumulator, and interruptible load power accumulator, respectively, each corresponding one-to-one with the four dimensions of the preset collaborative control target. The final values ​​of each dimension of a single representative decision feature vector are then compared with the target values ​​of the same dimension of the preset collaborative control target, and the absolute values ​​of each difference are calculated. The four absolute value results are then directly summed to obtain the total global absolute deviation corresponding to that representative decision feature vector.

[0067] Next, the sum of absolute deviations calculated from all representative decision feature vectors is uniformly aggregated. Using a basic ranking method of numerical comparison, the calculation result with the smallest global value is selected, and the representative decision feature vector corresponding to this result is identified. Its cluster set is then traced and designated as the positive preference cluster set. The cluster sets corresponding to all other representative decision feature vectors are uniformly designated as the negative preference cluster set, thus completing the classification labeling of clusters with different fit levels.

[0068] Finally, the intelligent fusion terminal pre-sets a threshold for the total number of preference data pairs. This threshold does not exceed the maximum number of samples that the intelligent fusion terminal can handle in a single round of computation. The sample extraction ratio between positive preference clusters and individual negative preference clusters is fixed at 1:1. The terminal iterates through all decision feature vectors stored in the positive preference clusters, using a random selection method and determining the extraction quantity according to the preset threshold. In each round, one decision feature vector is extracted as the winning data sample. Then, all the divided negative preference clusters are iterated through sequentially, and one decision feature vector is extracted from each negative preference cluster as the losing data sample, using the same random selection rule. Each winning sample is paired with a losing sample extracted from each negative preference cluster. After binding multiple sets of samples, they are uniformly collected and organized to form a complete set of response-consistent preference data pairs. The total number of data pairs in the set does not exceed the preset total number threshold.

[0069] By quantitatively calculating the multi-dimensional power deviation through clear dimensional mapping, the clustering levels are differentiated in terms of quality. Standardized preference data pairs are constructed by randomly pairing positive and negative samples with fixed extraction ratios and quantity constraints. This provides standardized, effective, and terminal-compatible basic data support for the iterative optimization of the collaborative control logic of energy equipment.

[0070] Furthermore, the method provided in this application embodiment includes: The initial values ​​of the four accumulators in the power accumulation engine and the power accumulation coefficients corresponding to each device type are used as trainable parameters. The preference data pair set is used as the supervision signal, and the pairwise ranking loss function is used to update the trainable parameters. The optimization objective is to make the final value of the accumulator output after the permutation sample sequence corresponding to the winner is input into the engine closer to the preset collaborative control objective, while making the final value of the accumulator corresponding to the loser far away from the preset collaborative control objective. When the standard deviation between the final values ​​of the accumulators of all decision feature vectors in the positive preference cluster set is less than two percent of their mean, training is stopped and the current trainable parameters are saved as the trained power accumulation engine parameters to obtain the trained power accumulation engine.

[0071] In one embodiment, the core operational logic of the power accumulation engine is first parameterized and decomposed. The initial values ​​of the four accumulators and the power accumulation coefficients corresponding to each device type code are set as trainable parameters. The four accumulators correspond to the photovoltaic power accumulator, the energy storage discharge power accumulator, the energy storage charging power accumulator, and the interruptible load power accumulator, respectively. The initial value of each accumulator is an independent floating-point trainable parameter with an initial default value of 0. The upper and lower limits of the parameter value are constrained from 0 to the maximum rated power of a single distributed energy device in the distribution area. The device type codes 01, 02, 03, and 04 each correspond to a set of independent power accumulation coefficients with an initial default value of 1. The upper and lower limits of the parameter value are constrained from 0.5 to 1.5. All trainable parameters are stored in the readable and writable storage area of ​​the terminal, supporting real-time reading and iterative updates during the training process.

[0072] Next, the constructed set of response consistency preference data pairs is used as the supervision signal for the training process. Each data pair contains a winning decision feature vector and a losing decision feature vector, corresponding to two sets of running state data sequences with different permutations. The training process uses a pairwise sorting loss function for iterative optimization. First, the permutation sample sequence corresponding to the winning side in the data pair is input into the power accumulation engine to be trained in its inherent order to obtain the corresponding accumulator final value, and the sum of the absolute deviations between the final value and the preset collaborative control target is calculated. Then, the permutation sample sequence corresponding to the losing side in the same data pair is input into the same power accumulation engine in its inherent order to obtain the corresponding accumulator final value, and the sum of the absolute deviations between the final value and the preset collaborative control target is calculated simultaneously. The calculation process of the pairwise sorting loss function is as follows: when the sum of the absolute deviations corresponding to the winner is greater than or equal to the sum of the absolute deviations corresponding to the loser, a positive loss value is generated; otherwise, the loss value is 0. The optimization goal of training is to continuously reduce the sum of the loss values ​​of all data pairs by iteratively updating the trainable parameters, so that the final value of the accumulator output by the engine after the permutation sample sequence corresponding to the winner is closer to the preset collaborative control target, while making the final value of the accumulator corresponding to the loser far away from the preset collaborative control target, and finally achieving that the output of the trained engine for different input sequences tends to be consistent.

[0073] After calculating the loss value for each set of preference data pairs, stochastic gradient descent is used to iteratively update all trainable parameters. During the update process, a fixed batch size of 8 and a weight decay coefficient of 1e-4 are set to avoid overfitting during training. Specifically, the update process involves calculating the partial derivative of the sum of the loss values ​​within a single batch with respect to each trainable parameter, adjusting the parameter values ​​according to a preset fixed learning rate. The learning rate ranges from 0.001 to 0.01 to adapt to the computational accuracy and capabilities of the intelligent fusion terminal. The parameter update process strictly adheres to pre-set upper and lower limits to prevent parameters from exceeding reasonable ranges. After each round of iterative training of the full set of preference data pairs, all decision feature vectors in the positive preference cluster set are traversed. The permutation sample sequence corresponding to each vector is sequentially input into the power accumulation engine after the current iteration. The final value of the accumulator corresponding to all samples is calculated. The standard deviation and mean of the final value of the accumulator under the four power dimensions are calculated respectively. When the standard deviation of the final value of the accumulator in all four dimensions is less than 2% of the mean of the corresponding dimension, the training is determined to have reached the convergence state. Iterative training is stopped immediately, and all the current trainable parameter values ​​are fixed and stored as the parameters of the power accumulation engine after training, thus completing the construction of the power accumulation engine after training.

[0074] By decomposing the power accumulation engine with explicit numerical constraints, using a pairwise sorting loss function driven by complete hyperparameter settings for preference optimization training, and employing a rigorous convergence determination mechanism, coupled with clear power control command conversion rules, the power accumulation engine achieves consistent output optimization for data with different input sequences. Ultimately, this effectively improves the stability of the intelligent fusion terminal's collaborative control of distributed energy devices and its alignment with grid dispatch objectives.

[0075] Furthermore, the method provided in this application embodiment includes: The multiple distributed energy devices include distributed photovoltaic inverters, energy storage converters, charging piles, and interruptible load circuit breakers.

[0076] Specifically, the four types of distributed energy devices correspond to the core power accumulation dimension of the power accumulation engine. Distributed photovoltaic inverters correspond to the photovoltaic power accumulation dimension, energy storage converters correspond to the dual accumulation dimension of energy storage charging power and energy storage discharging power, and charging piles and interruptible load circuit breakers correspond to the interruptible load power accumulation dimension. The operating status data of the four types of devices are directly input into the corresponding accumulator to complete the power accumulation calculation, realizing a one-to-one match between device type and engine calculation logic.

[0077] Meanwhile, the four types of distributed energy devices have differentiated asynchronous reporting timing characteristics. Distributed photovoltaic inverters report operating data frequently according to changes in light intensity, energy storage converters report operating data including charging and discharging status at fixed intervals, charging piles trigger high-frequency data reporting only when a charging load is connected, and interruptible load circuit breakers trigger event-based reporting only when the load switching status changes. These differentiated reporting characteristics of the four types of devices are the typical application scenario that urgently needs to be solved: "random data reporting order leads to inconsistent calculation results." The entire process of the aforementioned steps, including reordering, feature clustering, and engine optimization, can completely cover all possible combinations of reporting timing for the four types of devices.

[0078] After the intelligent fusion terminal completes the optimized training of the power accumulation engine, it generates corresponding collaborative control commands based on the stable and adjustable power output of the engine and sends them to four types of equipment. Specifically, it sends power adjustment commands to distributed photovoltaic inverters to control their active power output within the range of maximum and minimum adjustable power; it sends charge and discharge control commands to energy storage converters to adjust the charge and discharge power in combination with their state of charge; it sends charging power limit commands to charging piles to adjust their charging output; and it sends load switching commands to interruptible load circuit breakers to achieve precise control of interruptible loads, thereby completing closed-loop collaborative control of all equipment within the target control area.

[0079] In summary, the energy equipment collaborative control method based on intelligent fusion terminals provided in this application has the following technical effects: This application uses the power accumulation engine of the intelligent fusion terminal to process equipment operation data, generate decision feature vectors, and select representative vectors by clustering. Combined with response consistency preference data, it optimizes the training engine parameters, thereby eliminating the influence of data input order, making the collaborative control of distributed energy equipment more stable, improving the alignment with grid dispatch objectives, and achieving the technical effect that under the same physical state, the order of equipment reporting no longer affects the calculation results, and the terminal outputs a unique, stable, and accurate adjustable power.

[0080] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an energy equipment collaborative control system based on an intelligent fusion terminal, the system comprising: The permutation sample sequence generation module 1 is used to collect multiple operating status data of multiple distributed energy devices within the target control area, and to rearrange the multiple operating status data multiple times to generate multiple permutation sample sequences.

[0081] The decision feature vector acquisition module 2 is used to input multiple permutation sample sequences into the power accumulation engine built into the intelligent fusion terminal in sequence, extract the changes in the accumulator value inside the power accumulation engine when processing each permutation sample sequence and the final value of the accumulator after processing, and obtain multiple decision feature vectors.

[0082] Representative vector selection module 3 is used to traverse multiple decision feature vectors to perform unsupervised clustering, obtain multiple cluster sets, and select multiple representative decision feature vectors from the obtained multiple cluster sets respectively.

[0083] The preference data pair set construction module 4 is used to construct a response consistency preference data pair set based on the matching relationship between the multiple representative decision feature vectors and the preset collaborative control target.

[0084] The real-time collaborative control execution module 5 utilizes the response consistency preference data set to perform preference optimization training on the power accumulation engine, and deploys the trained power accumulation engine back to the intelligent fusion terminal to perform real-time collaborative control on the multiple distributed energy devices.

[0085] Furthermore, the permutation sample sequence generation module 1 is used to perform the following steps: If the number of multiple running state data is less than or equal to a first quantity threshold, then enumerate the factorial permutations of the multiple running state data and take each permutation as a permutation sample sequence to obtain multiple permutation sample sequences; otherwise, use a random permutation algorithm to generate a random permutation of a preset first value, and then select a permutation of a preset second value according to the principle of maximizing the number of element position swaps between permutations to obtain multiple permutation sample sequences, wherein the preset first value is greater than the preset second value.

[0086] Furthermore, the decision feature vector acquisition module 2 is used to perform the following steps: The power accumulation engine includes four accumulators: a photovoltaic power accumulator, an energy storage discharge power accumulator, an energy storage charging power accumulator, and an interruptible load power accumulator.

[0087] Furthermore, the decision feature vector acquisition module 2 is used to perform the following steps: After the power accumulation engine processes one running status data in each permutation sample sequence, it records the current values ​​of the four accumulators as a stage vector in the stage vector table of that permutation sample sequence. When the permutation sample sequence is finished, it concatenates the stage vector table of the permutation sample sequence with the final value of the accumulators of the permutation sample sequence to obtain the decision feature vector of the permutation sample sequence, and adds it to multiple decision feature vectors.

[0088] Furthermore, the decision feature vector acquisition module 2 is used to perform the following steps: The number of rows in the stage vector table is equal to the number of records of running state data in the permuted sample sequence, and each row contains four accumulator values.

[0089] Furthermore, the representative vector selection module 3 is used to perform the following steps: Calculate the Euclidean distance between every two decision feature vectors in multiple decision feature vectors to construct a distance matrix; treat each decision feature vector as an independent cluster, find the two independent clusters with the smallest Euclidean distance in the distance matrix and merge them to obtain a merged cluster; delete all Euclidean distances in the distance matrix related to the two independent clusters corresponding to the merged cluster, recalculate the Euclidean distances between the merged cluster and the remaining multiple decision feature vectors excluding the merged cluster, and update the distance matrix based on the calculation results; repeat agglomerative hierarchical clustering on the updated distance matrix until the maximum Euclidean distance within each merged cluster exceeds a preset threshold, then stop clustering to obtain multiple cluster sets.

[0090] Furthermore, the preference data set construction module 4 is used to perform the following steps: Obtain the target power value issued by the power grid dispatch automation system and use it as the preset collaborative control target; compare the final values ​​of the accumulators in multiple representative decision feature vectors with the preset collaborative control target, and calculate the sum of absolute deviations in four power dimensions; label the cluster set corresponding to the representative decision feature vector with the smallest sum of absolute deviations as the positive preference cluster set, and label the cluster sets corresponding to the remaining representative decision feature vectors as the negative preference cluster set; select one decision feature vector from the positive preference cluster set as the winner, and select one decision feature vector from the negative preference cluster set as the loser, and construct a response consistency preference data pair set.

[0091] Furthermore, the real-time collaborative control execution module 5 is used to perform the following steps: The initial values ​​of the four accumulators in the power accumulation engine and the power accumulation coefficients corresponding to each device type are used as trainable parameters. The preference data pair set is used as the supervision signal, and the pairwise ranking loss function is used to update the trainable parameters. The optimization objective is to make the final value of the accumulator output after the permutation sample sequence corresponding to the winner is input into the engine closer to the preset collaborative control objective, while making the final value of the accumulator corresponding to the loser far away from the preset collaborative control objective. When the standard deviation between the final values ​​of the accumulators of all decision feature vectors in the positive preference cluster set is less than two percent of their mean, training is stopped and the current trainable parameters are saved as the trained power accumulation engine parameters to obtain the trained power accumulation engine.

[0092] Furthermore, the permutation sample sequence generation module 1 is used to perform the following steps: The multiple distributed energy devices include distributed photovoltaic inverters, energy storage converters, charging piles, and interruptible load circuit breakers.

[0093] The energy equipment collaborative control system based on intelligent fusion terminal provided in the embodiments of the present invention can execute the energy equipment collaborative control method based on intelligent fusion terminal provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0094] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0095] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for coordinated control of energy equipment based on an intelligent fusion terminal, characterized in that, The method includes: Collect multiple operational status data from multiple distributed energy devices within the target control area, and rearrange the multiple operational status data multiple times to generate multiple permutation sample sequences; Multiple permutation sample sequences are sequentially input into the power accumulation engine built into the intelligent fusion terminal. The changes in the accumulator values ​​and the final value of the accumulator after processing each permutation sample sequence are extracted to obtain multiple decision feature vectors. Unsupervised clustering is performed by traversing multiple decision feature vectors to obtain multiple cluster sets, and multiple representative decision feature vectors are selected from the obtained multiple cluster sets respectively; Based on the matching relationship between the multiple representative decision feature vectors and the preset collaborative control target, a set of response consistency preference data pairs is constructed; Using the set of response consistency preference data pairs, preference optimization training is performed on the power accumulation engine. The trained power accumulation engine is then deployed back to the intelligent fusion terminal to perform real-time collaborative control of the multiple distributed energy devices. The process involves repeatedly rearranging the multiple operational status data to generate multiple permutation sample sequences, including: Determine whether the number of multiple running state data is less than or equal to the first quantity threshold. If so, enumerate the factorial permutations of the multiple running state data and take each permutation as a permutation sample sequence to obtain multiple permutation sample sequences. If not, a random permutation algorithm is used to generate a random permutation of the preset first value, and then permutations of the preset second value are selected according to the principle of maximizing the number of element position swaps between permutations, to obtain multiple permutation sample sequences, wherein the preset first value is greater than the preset second value.

2. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 1, characterized in that, The power accumulation engine includes four accumulators: a photovoltaic power accumulator, an energy storage discharge power accumulator, an energy storage charging power accumulator, and an interruptible load power accumulator.

3. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 2, characterized in that, Extracting the changes in accumulator values ​​within the power accumulation engine when processing each permutation sample sequence and the final value of the accumulator after processing, multiple decision feature vectors are obtained, including: After the power accumulation engine processes one running status data in each permutation sample sequence, it records the current values ​​of the four accumulators as a stage vector in the stage vector table of that permutation sample sequence. After the permutation sample sequence is processed, the stage vector table of the permutation sample sequence is concatenated with the final value of the accumulator of the permutation sample sequence to obtain the decision feature vector of the permutation sample sequence, and then added to multiple decision feature vectors.

4. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 3, characterized in that, The number of rows in the stage vector table is equal to the number of records of running state data in the permuted sample sequence, and each row contains four accumulator values.

5. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 1, characterized in that, Unsupervised clustering is performed by traversing multiple decision feature vectors to obtain multiple cluster sets, including: Calculate the Euclidean distance between every two decision feature vectors in a plurality of decision feature vectors, and construct a distance matrix; Each decision feature vector is treated as an independent cluster. The two independent clusters with the minimum Euclidean distance in the distance matrix are found and merged to obtain the merged cluster. Delete all Euclidean distances in the distance matrix related to the two independent clusters corresponding to the merged cluster, recalculate the Euclidean distances between the merged cluster and the remaining decision feature vectors excluding the merged cluster, and update the distance matrix based on the calculation results; Agglomerative hierarchical clustering is repeated on the updated distance matrix until the maximum Euclidean distance within each merged cluster exceeds a preset threshold, at which point clustering is stopped, resulting in multiple cluster sets.

6. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 1, characterized in that, Based on the matching relationship between the multiple representative decision feature vectors and the preset collaborative control objective, a set of response consistency preference data pairs is constructed, including: Obtain the target power value issued by the power grid dispatch automation system and use it as the preset collaborative control target; The final values ​​of the accumulators in multiple representative decision feature vectors are compared with the preset collaborative control target, and the sum of the absolute deviations in the four power dimensions is calculated. The cluster set corresponding to the representative decision feature vector with the smallest sum of absolute biases is labeled as the positive preference cluster set, and the cluster sets corresponding to the remaining representative decision feature vectors are labeled as the negative preference cluster set. Choose one decision feature vector from the positive preference cluster set as the winner and choose one decision feature vector from the negative preference cluster set as the loser to construct a set of response consistency preference data pairs.

7. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 1, characterized in that, Using the aforementioned response consistency preference data set, perform preference optimization training on the power accumulation engine, including: The initial values ​​of the four accumulators in the power accumulation engine and the power accumulation coefficients corresponding to each device type are used as trainable parameters. Using the set of preference data pairs as a supervision signal, the trainable parameters are updated using a pairwise sorting loss function. The optimization objective is to make the final value of the accumulator output by the engine after the permutation sample sequence corresponding to the winner is closer to the preset collaborative control objective, while making the final value of the accumulator corresponding to the loser far away from the preset collaborative control objective. When the standard deviation between the final values ​​of the accumulators of all decision feature vectors in the positive preference clustering set is less than two percent of their mean, training is stopped and the currently trainable parameters are saved as the parameters of the trained power accumulation engine, thus obtaining the trained power accumulation engine.

8. The energy equipment collaborative control method based on an intelligent fusion terminal as described in claim 1, characterized in that, The multiple distributed energy devices include distributed photovoltaic inverters, energy storage converters, charging piles, and interruptible load circuit breakers.

9. An energy equipment collaborative control system based on an intelligent fusion terminal, characterized in that, The system is used to implement the energy equipment collaborative control method based on an intelligent fusion terminal as described in any one of claims 1-8, the system comprising: The permutation sample sequence generation module is used to collect multiple operating status data of multiple distributed energy devices within the target control area, and to rearrange the multiple operating status data multiple times to generate multiple permutation sample sequences. The decision feature vector acquisition module is used to input multiple permutation sample sequences into the power accumulation engine built into the intelligent fusion terminal in sequence, extract the changes in the accumulator values ​​inside the power accumulation engine when processing each permutation sample sequence and the final value of the accumulator after processing, and obtain multiple decision feature vectors. The representative vector selection module is used to traverse multiple decision feature vectors to perform unsupervised clustering, obtain multiple cluster sets, and select multiple representative decision feature vectors from the obtained multiple cluster sets respectively; The preference data pair set construction module is used to construct a response consistency preference data pair set based on the matching relationship between the multiple representative decision feature vectors and the preset collaborative control target; The real-time collaborative control execution module uses the response consistency preference data set to perform preference optimization training on the power accumulation engine, and deploys the trained power accumulation engine back to the intelligent fusion terminal to perform real-time collaborative control on the multiple distributed energy devices.

Images