Distributed energy collaborative scheduling optimization method based on edge computing

Abstract

The invention discloses a distributed energy collaborative scheduling optimization method based on edge computing. According to the method, a plurality of edge computing nodes are deployed in a distributed energy system, a multi-protocol compatible OPC UA communication channel is constructed through protocol conversion middleware to collect data, and after the edge computing nodes clean and normalize the data, a preliminary scheduling scheme is generated through an improved genetic algorithm; the improved genetic algorithm is optimized through cooperation of a deep reinforcement learning model and an adaptive attenuation mechanism. And uploading the preliminary scheduling scheme to a cloud end, and obtaining a global optimal scheduling strategy through a multi-target particle swarm optimization algorithm. And the cloud carries out credible evidence storage on the global optimal scheduling strategy abstract value through an alliance chain smart contract, and establishes a PoA consensus mechanism. And when the communication is interrupted, the edge computing node starts the local emergency scheduling module, and incremental data synchronization is performed after the communication is recovered. The distributed energy scheduling optimization problem is effectively solved, the energy utilization efficiency is improved, and the system stability and reliability are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of distributed energy system optimization, and particularly to a distributed energy collaborative scheduling optimization method based on edge computing. Background Art

[0002] As an emerging computing mode, edge computing pushes computing tasks from the cloud to the network edge, enabling real-time data processing and analysis near the data source, effectively improving data processing efficiency and privacy protection. Applying edge computing to the collaborative scheduling of distributed energy systems can give full play to the advantages of edge computing and solve the problems existing in traditional scheduling methods. However, there are still some deficiencies in the current distributed energy collaborative scheduling methods based on edge computing:

[0003] In the dimension of data acquisition, the equipment brands and models in the distributed energy system are numerous and complex, and the data interfaces and transmission protocols of each device are like a mess, lacking a unified standard.

[0004] In the field of scheduling algorithms, the parameters of the local preliminary scheduling algorithm are fixed, which is prone to falling into local optimality, while it is difficult for the cloud global scheduling algorithm to take into account multiple objectives such as minimizing energy loss, node load balancing, and carbon emission reduction at the same time.

[0005] In terms of the security of data storage and verification, the security of the distributed energy system data is related to the stable operation and reliability of the energy supply network. The existing collaborative scheduling system based on edge computing lacks a complete encryption mechanism in the data storage link; in the verification link, the verification process and technical means are simple and crude; therefore, it is difficult to resist the ever-changing and emerging network attack means. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention proposes a distributed energy collaborative scheduling optimization method based on edge computing to solve at least one of the above technical problems.

[0007] The technical solution of the present invention is specifically as follows:

[0008] A distributed energy collaborative scheduling optimization method based on edge computing, the method comprising the following steps:

[0009] Step 1, deploy a plurality of edge computing nodes in the distributed energy system, and each edge computing node establishes a multi-protocol compatible OPC UA communication channel with a photovoltaic inverter, an energy storage battery management system, and a smart meter to construct a local data acquisition network;

[0010] Step 2: The edge computing node collects the photovoltaic power generation, the state of charge of the energy storage device, and the load power consumption data in real time, and integrates a data cleaning module in the edge computing node to clean and normalize the photovoltaic power generation, the state of charge of the energy storage device, and the load power consumption data;

[0011] Step 3: Run a local optimization module based on the improved genetic algorithm in each edge computing node to generate a preliminary scheduling plan that meets the local energy supply and demand balance. The improved genetic algorithm dynamically adjusts the crossover rate and mutation rate of the genetic algorithm through a deep reinforcement learning model collaborative adaptive attenuation mechanism;

[0012] Step 4: Upload the preliminary scheduling plans of each edge computing node to the cloud control center, and perform cross-node collaborative optimization through a multi-objective particle swarm optimization algorithm to generate a global optimal scheduling strategy. The optimization objectives include minimizing energy loss, balancing node loads, and reducing carbon emissions;

[0013] Step 5: The cloud control center stores the digest value of the global optimal scheduling strategy in a trusted manner through the consortium chain smart contract. Each edge computing node, as a lightweight blockchain node, uses an improved PoA consensus mechanism based on reputation scores for hash verification;

[0014] Step 6: When the communication between the edge computing node and the cloud control center is interrupted for more than a preset threshold, the edge computing node starts the local emergency scheduling module to automatically switch to the dynamic time window historical optimal scheduling plan stored in the local database, and performs incremental data synchronization through differential hash comparison after the communication is restored.

[0015] Further, in Step 1, the OPC UA communication channel realizes multi-protocol compatibility through a protocol conversion middleware. The protocol conversion middleware includes a transmission protocol parsing unit, a data standardization encapsulation unit, and an OPC UA server interface unit;

[0016] The transmission protocol parsing unit is used to convert the original data of the photovoltaic inverter and the energy storage battery management system into a structured JSON data stream;

[0017] The data standardization encapsulation unit is used to add a device identifier and a timestamp to the JSON data stream. The device identifier includes a device type code and a geographical location code;

[0018] The OPC UA server interface unit is used to publish the encapsulated data as an OPC UA node.

[0019] Further, in Step 3, the specific formulas for the adaptive attenuation mechanism to dynamically adjust the crossover rate and mutation rate of the genetic algorithm are:

[0020]

[0021] In formula (1), P c0 , P m0 are the initial crossover rate and mutation rate, α and β are attenuation coefficients, and t is the number of iterations.

[0022] Furthermore, the attenuation coefficient is dynamically adjusted according to the population diversity:

[0023] When the population diversity index is lower than the preset threshold, increase the attenuation coefficient to accelerate convergence, and increase the coefficient γ inc ∈[1.1, 2.0];

[0024] When the population diversity index is higher than the preset threshold, decrease the attenuation coefficient to maintain the search range, and decrease the coefficient γ dec ∈[0.5, 0.9].

[0025] Furthermore, in step 3, the deep reinforcement learning model adopts a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm model, which is used to optimize the initial values of the crossover rate and mutation rate of the genetic algorithm and the benchmark value of the attenuation coefficient. The state space, action space, and reward function of the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm are as follows:

[0026] The state space:

[0027] s t = [S f , t, ΔE](2)

[0028] In formula (2), S f is the variance of the population fitness, t is the number of iterations, and ΔE is the energy supply-demand deviation;

[0029] The action space is:

[0030] a t = [ΔP c0 , ΔP m0 , Δα base , Δβ base (3)

[0031] In formula (3), ΔP c0 is the adjustment amount of the initial crossover rate, ΔP m0 is the adjustment amount of the initial mutation rate, Δα base and Δβ base are the adjustment amounts of the benchmark value of the attenuation coefficient;

[0032] The reward function is:

[0033] R = η1·(f prev - f current) + η2·D(4)

[0034] In formula (4), f prev and f current are the optimal fitness values of the previous and current generations, D is the population diversity index, and η1 and η2 are weight coefficients.

[0035] Furthermore, the specific method for the double delayed deep deterministic policy gradient (TD3) algorithm model to dynamically adjust the crossover rate and mutation rate of the genetic algorithm through the co - adaptive decay mechanism is as follows:

[0036] S1. Before each iteration of the genetic algorithm, the double delayed deep deterministic policy gradient (TD3) algorithm model updates the initial values of the crossover rate and mutation rate and the reference values of the decay coefficients:

[0037]

[0038] In formulas (5) and (6), t is the number of iterations, and are the updated initial values of the crossover rate and mutation rate, and are the current initial values of the crossover rate and mutation rate, ΔP c0 and ΔP m0 are the adjustment amounts of the crossover rate and mutation rate, and are the updated reference values of the decay coefficients, and are the current reference values of the decay coefficients, and are the adjustment amounts of the reference values of the decay coefficients;

[0039] S2. During the operation of the genetic algorithm, the actual decay coefficient is dynamically corrected through the population diversity index:

[0040]

[0041] In formulas (7) and (8), α (t) and β (t) are the adjusted decay coefficients, and are the reference values of the decay coefficients, D is the population diversity index, D threshold is the population diversity threshold, γ inc and γ dec are the adjustment coefficients, and γ inc >1, γ dec <1;

[0042] S3. Calculate the actual crossover rate and mutation rate according to the updated parameters through the adaptive decay mechanism:

[0043]

[0044] In formula (9), P c and P m are the updated crossover rate and mutation rate, and are the updated crossover rate and mutation rate, α (t) , β (t) are the updated attenuation coefficients, and t is the number of iterations.

[0045] Furthermore, in step 4, the objective function of the multi-objective particle swarm optimization algorithm is:

[0046] min{w1·f loss +w2·f load +w3·f carbon}(10)

[0047] In formula (10), f loss is the energy transmission loss, f load is the node load imbalance index, f carbon is the carbon emission, w1, w2 and w3 are weight coefficients, and satisfy w1 + w2 + w3 = 1.

[0048] Furthermore, in step 5, the summary value of the global optimal scheduling strategy is generated through a Merkle tree structure, and the consortium chain smart contract executes the following steps:

[0049] L1. Split the global optimal scheduling strategy into multiple data blocks through dynamic priorities, and add priority tags to the headers of the data blocks;

[0050] L2. Calculate the hash values of each data block hierarchically according to the priority tags and construct a Merkle tree. When the scheduling strategy is locally updated, only the hash values of the affected data blocks and paths are recalculated;

[0051] L3. The edge computing nodes use the BLS threshold signature algorithm to generate a distributed signature. After binding the signature data with the Merkle root hash, the summary value of the global optimal scheduling strategy is written into the smart contract to trigger a deposit event;

[0052] L4. Real-time monitor the system status through the smart contract rule engine, and trigger an emergency deposit when the load suddenly changes or the battery level of the energy storage device is too low;

[0053] L5. Pre-compute the Merkle verification path of each data block in the cloud control center and compress and store it. Each edge computing node caches the path after the first verification.

[0054] Further, in the step 5, the improved PoA consensus mechanism includes:

[0055] H1. Calculate the reputation scores of each edge computing node. The formula for calculating the reputation score of the edge computing node is:

[0056]

[0057] In formula (11), N verify is the number of successful verifications, T response is the response time, A data is the data accuracy rate, and λ1, λ2, and λ3 are weight coefficients, and satisfy λ1 + λ2 + λ3 = 1;

[0058] H2. Perform verification grading through reputation scores:

[0059] When S ≥ 80, the edge computing node is a high-reputation node and is responsible for block generation and signature;

[0060] When 60 ≤ S < 80, the edge computing node is an alternative node and only participates in hash verification and has no right to generate blocks or signatures;

[0061] When S < 60, the edge computing node is a low-reputation node, and its verification permission is suspended and needs to be restored after manual review;

[0062] H3. The smart contract generates a new list of verification nodes based on the latest reputation scores at the end of each 24-hour cycle;

[0063] H4. When multiple edge computing nodes submit the same incorrect data within a short period of time, trigger collusion detection and determine the nodes as colluding nodes. The integral of the colluding nodes is cleared and they are removed from the verification list for at least 48 hours.

[0064] Further, in the step 6, the generation method of the dynamic time window historical optimal scheduling scheme includes:

[0065] K1. Dynamically match the time window length according to the historical weather type, load pattern, and energy storage SOC interval;

[0066] K2. Use a sliding window to calculate the correlation coefficient between the current data and historical data, and filter out the historical periods whose correlation coefficient exceeds the set threshold;

[0067] K3. Screen the top K optimal scheduling schemes from the historical periods by calculating the Euclidean distance, and generate an emergency plan after weighted averaging.

[0068] The beneficial effects of the present invention are as follows: Through the multi - protocol - compatible OPC UA communication channel and the protocol conversion middleware, the problem of inconsistent device data interfaces and transmission protocols in the distributed energy system is effectively solved, ensuring the integrity and accuracy of data collection.

[0069] The improved genetic algorithm dynamically adjusts parameters through the collaborative adaptive attenuation mechanism of the deep reinforcement learning model, avoiding falling into local optima and improving the global search ability of the algorithm. The multi - objective particle swarm optimization algorithm can take into account multiple objectives such as minimizing energy loss, balancing node load, and reducing carbon emissions at the same time.

[0070] Based on the trusted deposit of the consortium chain smart contract and the verification of the improved PoA consensus mechanism, the security and immutability of the scheduling policy data are ensured, effectively resisting network attacks and greatly reducing the risk of data tampering. The emergency scheduling scheme and the incremental data synchronization mechanism after communication recovery ensure the stable operation and data consistency of the system in case of communication anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is the overall flowchart of a distributed energy collaborative scheduling optimization method based on edge computing according to the present invention;

[0072] Figure 2 is the detailed flowchart of the collaborative adaptive attenuation mechanism of the (TD3) algorithm model according to the present invention;

[0073] Figure 3 is the detailed flowchart of the consortium chain smart contract according to the present invention;

[0074] Figure 4 is the detailed flowchart of the improved PoA consensus mechanism according to the present invention;

[0075] Figure 5 is the detailed flowchart of generating the historical optimal scheduling scheme with a dynamic time window according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative work belong to the protection scope of the present invention.

[0077] Please refer to Figures 1 to 5 , the present invention provides a distributed energy collaborative scheduling optimization method based on edge computing, including the following steps:

[0078] Step 1: Deploy multiple edge computing nodes in the distributed energy system. Each edge computing node builds a local data acquisition network by establishing an OPC UA communication channel compatible with multiple protocols with the photovoltaic inverter, energy storage battery management system, and smart meter.

[0079] Step 2: The edge computing nodes collect real-time data on photovoltaic power generation, the state of charge of energy storage devices, and load electricity demand. And an integrated data cleaning module in the edge computing nodes cleans and normalizes the data on photovoltaic power generation, the state of charge of energy storage devices, and load electricity demand.

[0080] Step 3: Run a local optimization module based on the improved genetic algorithm in each edge computing node to generate a preliminary scheduling plan that meets the local energy supply and demand balance. The improved genetic algorithm dynamically adjusts the crossover rate and mutation rate of the genetic algorithm through a deep reinforcement learning model collaborative adaptive attenuation mechanism.

[0081] Step 4: Upload the preliminary scheduling plans of each edge computing node to the cloud control center, and perform cross-node collaborative optimization through a multi-objective particle swarm optimization algorithm to generate a global optimal scheduling strategy. The optimization objectives include minimizing energy loss, balancing node loads, and reducing carbon emissions.

[0082] Step 5: The cloud control center stores the summary value of the global optimal scheduling strategy in a trusted manner through the consortium chain smart contract. Each edge computing node, as a lightweight blockchain node, performs hash verification using an improved PoA consensus mechanism based on reputation scores.

[0083] Step 6: When the communication between the edge computing node and the cloud control center is interrupted for more than a preset threshold, the edge computing node starts the local emergency scheduling module to automatically switch to the dynamic time window historical optimal scheduling plan stored in the local database, and performs incremental data synchronization through differential hash comparison after the communication is restored.

[0084] Specifically, Step 1 is the deployment of edge computing nodes and the construction of the OPC UA communication channel for the distributed energy collaborative scheduling optimization method based on edge computing. Edge computing nodes are deployed in photovoltaic power plants, energy storage stations, and load centers. The edge computing node uses an ARM Cortex-A72 quad-core processor, supports parallel processing of multi-protocol data parsing, has a memory of 8GB DDR4, supports running data acquisition, cleaning, and local optimization algorithms simultaneously, and has a storage of 128GB eMMC to store historical data and the emergency scheduling plan library.

[0085] In Step 1, the OPC UA communication channel realizes multi-protocol compatibility through a protocol conversion middleware. The protocol conversion middleware includes a transmission protocol parsing unit, a data standardization encapsulation unit, and an OPC UA server interface unit.

[0086] The transmission protocol parsing unit is used to convert the original data of the photovoltaic inverter and the energy storage battery management system into a structured JSON data stream;

[0087] Specifically, the transmission protocol parsing unit supports Modbus RTU, CAN bus, and DLT645 protocols;

[0088] The data standardization and encapsulation unit is used to add a device identifier and a timestamp to the JSON data stream, and the device identifier includes a device type code and a geographical location code;

[0089] Specifically, the geographical location code adopts the GB / T 2260 standard, the first 6 digits are the administrative division code, and the format of the device identifier is: {device type code}_{geographical location code}_{sequence number}. For example, PV_420111_001 represents the No. 1 photovoltaic inverter in Hongshan District, Wuhan City; the timestamp is synchronized to the UTC+8 time zone using the NTP protocol, with an accuracy of ±1ms.

[0090] The OPC UA server interface unit is used to publish the encapsulated data as OPC UA nodes.

[0091] Specifically, an OPC UA modeling tool is used to create a device type template to dynamically generate nodes, support periodic publishing and event-triggered publishing, and use TimescaleDB to store historical data, supporting queries by time range.

[0092] After unifying the data format, the subsequent algorithm processing efficiency is increased by 50%, and the data conflict rate is reduced from 1.2% to 0.1%. When adding support for new protocols, only the corresponding parsing plug-in needs to be developed without modifying the core code.

[0093] Specifically, step 2 is the data collection, data cleaning, and normalization processing of the distributed energy collaborative scheduling optimization method based on edge computing. Data is collected through sensors deployed at the photovoltaic power station, energy storage station, and load center. Current sensors, voltage sensors, and irradiance sensors are deployed at the photovoltaic power station; SOC sensors and temperature sensors are deployed at the energy storage station; smart meters and demand transmitters are deployed at the load center; the edge computing node integrates a data cleaning module to clean and normalize the photovoltaic power generation power, energy storage state of charge, and load power consumption demand data.

[0094] Specifically, the data cleaning module identifies abnormal data through an anomaly detection algorithm based on a sliding window Z-score. After marking the abnormal data as abnormal, the missing values are filled by cubic spline interpolation. The data normalization processing includes Min-Max normalization processing and Z-score standardization processing.

[0095] Specifically, the sliding window Z-score includes:

[0096] Set the window length to 10 minutes, which includes 600 sampling points. Calculate the mean μ and standard deviation σ of the sampling points. The anomaly judgment threshold is:

[0097] |x - μ| > 3σ

[0098] where x is the value of the sampling point;

[0099] Dynamically adjust the window size:

[0100] Photovoltaic power fluctuation detection: When the standard deviation of 3 consecutive windows > 0.2, the window is shortened to 5 minutes;

[0101] Load demand stable period: When the standard deviation of 5 consecutive windows < 0.05, the window is extended to 15 minutes.

[0102] Specifically, step 3 is a distributed energy collaborative scheduling optimization method based on edge computing. At the edge computing node, a preliminary scheduling plan that meets the local energy supply and demand balance is generated through an improved genetic algorithm. Each chromosome of the improved genetic algorithm represents a scheduling plan for 24 hours, and each hour contains 3 decision variables. The decision variables are the supply-demand balance rate B, the energy storage life loss L, and the load fluctuation C. Therefore, the fitness function of the improved genetic algorithm is designed as:

[0103] f = τ1·B + τ2·L + τ3·C

[0104] The optimal scheduling plan that meets the local energy supply and demand balance is obtained through selection, crossover, and mutation operations as the preliminary scheduling plan.

[0105] The improved genetic algorithm dynamically adjusts the crossover rate and mutation rate of the genetic algorithm through a deep reinforcement learning model collaborative adaptive decay mechanism, enabling the crossover rate and mutation rate to be dynamically adjusted according to real-time data characteristics. Compared with the traditional fixed parameter method, the optimization efficiency is increased by 40%.

[0106] In step 3, the specific formulas for the adaptive decay mechanism to dynamically adjust the crossover rate and mutation rate of the genetic algorithm are:

[0107]

[0108] In formula (1), P c0 、P m0 are the initial crossover rate and mutation rate, α and β are the decay coefficients, and t is the number of iterations.

[0109] The decay coefficient is dynamically adjusted according to the population diversity:

[0110] When the population diversity index is lower than the preset threshold, increase the attenuation coefficient to accelerate convergence and increase the coefficient γ inc ∈ [1.1, 2.0];

[0111] When the population diversity index is higher than the preset threshold, decrease the attenuation coefficient to maintain the search range and decrease the coefficient γ dec ∈ [0.5, 0.9].

[0112] In step 3, the deep reinforcement learning model adopts a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm model, which is used to optimize the initial values of the genetic algorithm crossover rate and mutation rate and the attenuation coefficient reference value. The state space, action space, and reward function of the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm are as follows:

[0113] The state space:

[0114] s t = [S f , t, ΔE](2)

[0115] In formula (2), S f is the population fitness variance, t is the number of iterations, and ΔE is the energy supply - demand deviation;

[0116] The action space is:

[0117] a t = [ΔP c0 , ΔP m0 , Δα base , Δβ base (3)

[0118] In formula (3), ΔP c0 is the initial crossover rate adjustment amount, ΔP m0 is the initial mutation rate adjustment amount, Δα base and Δβ base are the attenuation coefficient reference value adjustment amounts;

[0119] The reward function is:

[0120] R = η1·(f prev - f current ) + η2·D(4)

[0121] In formula (4), f prev and f current are the optimal fitness of the previous and next generations, D is the population diversity index, and η1 and η2 are weight coefficients.

[0122] The specific method for the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm model to dynamically adjust the genetic algorithm crossover rate and mutation rate through a collaborative adaptive attenuation mechanism is as follows:

[0123] S1. Before each iteration of the genetic algorithm, the initial values and decay coefficient reference values of the crossover rate and mutation rate are updated in the double delayed deep deterministic policy gradient (TD3) algorithm model:

[0124]

[0125] In formulas (5) and (6), t is the number of iterations, and are the updated initial values of the crossover rate and mutation rate, and are the current initial values of the crossover rate and mutation rate, ΔP c0 and ΔP m0 are the adjustment amounts of the crossover rate and mutation rate, and are the updated decay coefficient reference values, and are the current decay coefficient reference values, and are the adjustment amounts of the decay coefficient reference value;

[0126] S2. During the operation of the genetic algorithm, the actual decay coefficient is dynamically corrected through the population diversity index:

[0127]

[0128] In formulas (7) and (8), α (t) and β (t) are the adjusted decay coefficients, and are the decay coefficient reference values, D is the population diversity index, D threshold is the population diversity threshold, γ inc and γ dec are the adjustment coefficients, and γ inc > 1, γ dec < 1;

[0129] S3. Calculate the actual crossover rate and mutation rate according to the updated parameters through the adaptive decay mechanism:

[0130]

[0131] In formula (9), P c and P m are the updated crossover rate and mutation rate, and are the updated crossover rate and mutation rate, α (t) and β (t) are the updated decay coefficients, and t is the number of iterations.

[0132] Specifically, step 4 is the cross-node collaborative optimization of the distributed energy collaborative scheduling optimization method based on edge computing through the multi-objective particle swarm optimization algorithm to generate the global optimal scheduling strategy. The optimization objectives include minimizing energy loss, balancing node load, and reducing carbon emissions;

[0133] In step 4, the objective function of the multi-objective particle swarm optimization algorithm is:

[0134] min{w1·f loss +w2·f load +w3·f carbon}(10)

[0135] In formula (10), f loss is the energy transmission loss, f load is the node load imbalance index, f carbon is the carbon emission, w1, w2, and w3 are weight coefficients, and satisfy w1 + w2 + w3 = 1.

[0136] Specifically, the energy transmission loss is calculated as: where I i is the line current, R i is the line resistance, and Δt is the time interval;

[0137] The load imbalance index is calculated as: where P i is the node load, is the average load;

[0138] The carbon emission is calculated as: where is the electricity purchased from the power grid, ρ grid = 0.38 kg / kWh is the carbon emission factor of thermal power, is the gas power generation, ρ gas = 0.21 kg / kWh is the carbon emission factor of natural gas;

[0139] The constraint conditions are: energy storage capacity constraint: 0 ≤ SOC k ≤ SOC max,k , where SOC k is the state of charge of the energy storage device, and SOC max,k is the maximum available capacity of the energy storage device;

[0140] Device output constraint: P i min ≤ P i ≤ P i max , where P i is the active output of the device, Pi min is the minimum output of the device, P i max is the maximum output of the device;

[0141] Power balance constraint: where G is the total number of power generation devices, P i is the output of the power generation device, L is the total number of load nodes, W j is the demand of the load node.

[0142] Specifically, step 5 is the blockchain trusted deposit of the distributed energy collaborative scheduling optimization method based on edge computing. The consortium blockchain includes a master node and lightweight nodes. The master node is the cloud control center, responsible for generating blocks and verifying signatures. The lightweight nodes are edge computing nodes, only storing Merkle paths and verifying signatures. The smart contract platform is Hyperledger Fabric 2.4, supporting the development of chaincode. The deposit contract function of the smart contract platform includes receiving the summary value of the scheduling policy, performing BLS signature aggregation, triggering Merkle tree update, and recording deposit events.

[0143] In step 5, the summary value of the global optimal scheduling policy is generated through the Merkle tree structure, and the consortium chain smart contract executes the following steps:

[0144] L1. Split the global optimal scheduling policy into multiple data blocks through dynamic priority and add priority tags to the headers of the data blocks;

[0145] Specifically, the priorities are divided into:

[0146] High priority (P1): Energy storage charge and discharge instructions, load shedding commands;

[0147] Medium priority (P2): Prediction data, device status;

[0148] Low priority (P3): Historical data, log information;

[0149] L2. According to the priority tags, calculate the hash values of each data block hierarchically and construct a Merkle tree. When the scheduling policy is locally updated, only recalculate the hash values of the affected data blocks and paths;

[0150] L3. The edge computing node uses the BLS threshold signature algorithm to generate a distributed signature. After binding the signature data with the Merkle root hash, generate the summary value of the global optimal scheduling policy and write it into the smart contract, triggering a deposit event;

[0151] Specifically, the master node generates a private key s, the public key P = s·G, and the edge computing node generates a partial private key Si , and sent to the master node;

[0152] L4. Real-time monitor the system status through the smart contract rule engine, and trigger emergency evidence storage when the load suddenly changes or the power of the energy storage device is too low;

[0153] L5. Pre-compute the Merkle verification path for each data block in the cloud control center and compress and store it. Each edge computing node caches the path after the first verification.

[0154] In the step 5, the improved PoA consensus mechanism includes:

[0155] H1. Calculate the reputation scores of each edge computing node. The formula for calculating the reputation score of the edge computing node is:

[0156]

[0157] In formula (11), N verify is the number of successful verifications, T response is the response time, A data is the data accuracy rate, and λ1, λ2, and λ3 are weight coefficients, and satisfy λ1 + λ2 + λ3 = 1;

[0158] H2. Perform verification grading through reputation scores:

[0159] When S ≥ 80, the edge computing node is a high-reputation node and is responsible for block generation and signature;

[0160] When 60 ≤ S < 80, the edge computing node is an alternative node and only participates in hash verification and has no right to generate blocks or signatures;

[0161] When S < 60, the edge computing node is a low-reputation node, and the verification permission is suspended and needs to be restored after manual review;

[0162] H3. The smart contract generates a new list of verification nodes according to the latest reputation scores at the end of each 24-hour cycle;

[0163] H4. When multiple edge computing nodes submit the same error data in a short period of time, trigger collusion detection to determine that the nodes are colluding nodes. The scores of the colluding nodes are cleared and they are removed from the verification list for at least 48 hours.

[0164] Specifically, step 6 is the emergency scheduling and data synchronization of the distributed energy collaborative scheduling optimization method based on edge computing. The response condition of the emergency strategy is that the communication interruption threshold is 5 minutes and can be configured, and the emergency scheduling is triggered when the energy storage SOC threshold is lower than 20%. The priority of the local plan is:

[0165] High priority: The real-time optimal plan in the recent 1 hour;

[0166] Medium priority: For historical solutions of the same type, the types are weather and load matching;

[0167] Low priority: The guaranteed solution is that energy storage supplies power first, and the load is reduced by 30%;

[0168] In the step 6, the method for generating the dynamic time window historical optimal scheduling solution includes:

[0169] K1. Dynamically match the time window length according to the historical weather type, load pattern, and energy storage SOC interval;

[0170] K2. Use a sliding window to calculate the correlation coefficient between the current data and historical data, and filter out historical periods with a correlation coefficient exceeding the set threshold;

[0171] K3. Screen the top K optimal scheduling solutions from the historical periods by calculating the Euclidean distance, and generate an emergency solution after weighted averaging.

[0172] There are multiple manufacturing enterprises in an industrial park in Zhenping County, with large and obvious fluctuations in electricity demand. The park is equipped with a distributed energy system, including 5MWp of photovoltaic power generation facilities, consisting of more than 4,000 1250W photovoltaic modules, distributed on the rooftops of factory buildings in the park; 2MW of wind power generation equipment, equipped with 4 500kW wind turbines, installed in the open area of the park; and a 3MWh lithium battery energy storage system, placed in a dedicated energy storage station. At the same time, an 800kW gas turbine is also set as a supplementary power source. The loads in the park mainly include industrial production equipment with a peak power of up to 8MW, and public facilities such as air conditioners and lighting with a power of about 1.5MW. By applying the distributed energy collaborative scheduling optimization method based on edge computing in the park's energy system, the energy output can be greatly stabilized, the efficiency of energy collaborative scheduling can be improved, and resource waste can be reduced.

[0173] Deploy edge computing nodes at each energy facility and electrical load in the park:

[0174] Photovoltaic array deployment: One high-performance edge computing device is configured for every 1MW photovoltaic area, connected to 20 inverters through the RS485 bus to collect data such as the voltage and current of photovoltaic modules in real time.

[0175] Wind turbine deployment: An industrial-grade edge computing node is deployed inside the tower of each wind turbine, integrated with a high-precision wind speed sensor, and communicates with the wind turbine converter through the CAN bus to obtain information such as the wind turbine speed and power generation power.

[0176] Energy storage station deployment: Install an edge computing module beside the battery management system of the energy storage station, support the IEC 61850 standard protocol, and realize the monitoring of the voltage, charge and discharge current, and remaining power of the energy storage battery.

[0177] Deployment in the power distribution room: Deploy edge computing gateways in the power distribution room to connect smart meters and collect power consumption data for each area.

[0178] Each edge computing node communicates with the cloud control center of the park through the 5G network to ensure fast data transmission and interaction.

[0179] Edge computing nodes collect data on photovoltaic, wind power, energy storage, and loads in real time at a frequency of 1Hz. For the collected photovoltaic power data, on sunny days, the power is close to 5MW, while on cloudy days, it drops to about 3MW. The wind power varies according to the wind speed. When the wind speed is 10m / s, the power generation can reach 1.5MW. The sliding window Z-score algorithm is used to detect anomalies in the data. When the photovoltaic power drops suddenly by more than 20% in a short period of time, it is marked as an abnormal data point. Cubic spline interpolation is used to fill in the missing values. All data is normalized, and data such as photovoltaic power, energy storage SOC, and load demand are uniformly mapped to the [0,1] interval for convenient subsequent analysis and calculation.

[0180] An improved genetic algorithm is used to generate a preliminary scheduling plan at the edge computing node. The 24-hour energy scheduling is encoded, and the decision variables include the hourly photovoltaic output (0 - 5MW), the energy storage charge and discharge amount (-3MWh to 3MWh), and the start / stop status of the gas turbine. The fitness function comprehensively considers the energy supply-demand balance rate, energy usage cost, and carbon emissions, where the weight of the supply-demand balance rate is 0.4, the energy cost weight is 0.3, and the carbon emission weight is 0.3. The TD3 algorithm is used to dynamically adjust the crossover rate and mutation rate of the genetic algorithm. When the population diversity is low, the mutation rate is appropriately increased to increase the population diversity. After 50 iterations of optimization, a scheduling plan that meets the local energy supply-demand balance is initially obtained, with the photovoltaic consumption rate reaching 85% and the energy storage cycle times controlled within 4 times per day.

[0181] Upload the preliminary plan generated by the edge computing node to the cloud control center, and the cloud control center performs global collaborative optimization based on the multi-objective particle swarm optimization algorithm. The objective function consists of the weighted sum of transmission loss, load balance, and carbon emissions, and the weights are set to 0.5, 0.3, and 0.2 respectively. During the optimization process, considering the characteristics of the power transmission lines in each area of the park, the power transmission loss is reduced by adjusting the energy distribution plan, from 300kW to less than 200kW. At the same time, the load distribution is optimized, and the load balance index is reduced from 0.7 to about 0.4 to ensure stable power consumption in each area. In terms of carbon emissions, by reasonably allocating energy, preferentially using clean energy, and reducing the usage time of gas turbines, the carbon emissions are reduced by 20%. In addition, the energy interaction between the park and the surrounding areas is also considered. When there is an excess of electricity, the surplus photovoltaic power is transmitted to the surrounding areas, improving the energy utilization efficiency and realizing the cross-regional optimal allocation of energy.

[0182] Use blockchain technology to conduct trustworthy deposit of the scheduling plan and energy data to ensure the authenticity, immutability, and traceability of the data. Take key energy scheduling instructions, such as the start-stop instructions of gas turbines and the charge-discharge control instructions of energy storage, as high-priority data blocks, and use the SHA-256 hash algorithm to generate hash values. Organize them according to the data structure of the Merkle tree, and gradually merge and calculate the hash values of multiple data blocks to finally generate the root hash value. Multiple edge computing nodes in the park participate in the BLS threshold signature. Each node has a partial private key. Through the distributed signature mechanism, the signature is only valid after a certain number of node signatures are satisfied, ensuring the security and credibility of the data during transmission and storage, and effectively preventing the data from being maliciously tampered with.

[0183] When the park encounters emergencies such as communication interruption and sudden equipment failure, start the emergency scheduling mechanism. Match the historical data through a dynamic time window, and select the historical scheduling plan under similar working conditions as the emergency plan according to factors such as the current energy supply and demand situation and meteorological conditions. For example, when a sudden cloudy day causes a sharp reduction in photovoltaic power generation, the system will quickly match the scheduling plan in the past when it was cloudy and the load was similar. Use the differential hash synchronization mechanism to quickly compare the differences between the local data and the cloud data after the communication is restored, and only transmit the different parts of the data to achieve rapid data synchronization, ensure the stable operation of the energy system, and minimize the impact of emergencies on the production and operation of the park.

[0184] By applying the distributed energy collaborative scheduling optimization method based on edge computing of the present invention to the park energy system, the comparison of the results before and after the optimization of the park energy scheduling is shown in Table 1:

[0185] Table 1 Comparison of the results before and after the optimization of the park energy scheduling

[0186] Indicator Before optimization After optimization Improvement rate Photovoltaic accommodation rate 75％ 88％ +13％ Energy storage cycle times 5 times per day 4 times per day -20％ Demand charge 9.3 yuan per kilowatt-hour 7.6 yuan per kilowatt-hour -18％ Carbon emissions 12 tons per day 9.36 tons per day -22％ Power supply interruption time 20 minutes 8 seconds -99.3％

[0187] As can be seen from the above embodiments, the distributed energy collaborative scheduling optimization method based on edge computing according to the present invention has significant technical effects and improves the energy utilization efficiency of the distributed energy system.

[0188] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present application. In addition, the specific details disclosed above are only for illustrative and easy-to-understand purposes and are not limitations. The above details do not limit the present application to necessarily adopt the above specific details for implementation.

[0189] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present application are only illustrative examples and do not intend to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open-ended words, meaning "including but not limited to", and can be used interchangeably with each other. The words "or" and "and" used herein refer to the word "and / or" and can be used interchangeably with each other, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with each other.

[0190] It should also be noted that in the devices, equipment, and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present application.

[0191] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be very apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0192] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A distributed energy collaborative scheduling optimization method based on edge computing, characterized in that The method includes the following steps: Step 1: Deploy multiple edge computing nodes in the distributed energy system. Each edge computing node builds a local data acquisition network by establishing an OPC UA communication channel compatible with multiple protocols with the photovoltaic inverter, energy storage battery management system, and smart meter; Step 2: The edge computing node collects the photovoltaic power generation, the state of charge of the energy storage device, and the load power consumption demand data in real time, and integrates a data cleaning module in the edge computing node to clean and normalize the photovoltaic power generation, the state of charge of the energy storage device, and the load power consumption demand data; Step 3: Run a local optimization module based on the improved genetic algorithm in each edge computing node to generate a preliminary scheduling plan that meets the local energy supply and demand balance. The improved genetic algorithm dynamically adjusts the crossover rate and mutation rate of the genetic algorithm through a deep reinforcement learning model collaborative adaptive attenuation mechanism; Step 4: Upload the preliminary scheduling plans of each edge computing node to the cloud control center, and perform cross-node collaborative optimization through a multi-objective particle swarm optimization algorithm to generate a global optimal scheduling strategy. The optimization objectives include minimizing energy loss, balancing node loads, and reducing carbon emissions; Step 5: The cloud control center stores the digest value of the global optimal scheduling strategy in a trusted manner through the consortium blockchain smart contract. Each edge computing node, as a lightweight blockchain node, performs hash verification using an improved PoA consensus mechanism based on reputation scores; Step 6: When the communication between the edge computing node and the cloud control center is interrupted for more than a preset threshold, the edge computing node starts the local emergency scheduling module to automatically switch to the dynamic time window historical optimal scheduling plan stored in the local database, and performs incremental data synchronization through differential hash comparison after the communication is restored.

2. The distributed energy collaborative scheduling optimization method based on edge computing according to claim 1, wherein: In Step 1, the OPC UA communication channel realizes multi-protocol compatibility through a protocol conversion middleware. The protocol conversion middleware includes a transmission protocol parsing unit, a data standardization encapsulation unit, and an OPC UA server interface unit; The transmission protocol parsing unit is used to convert the original data of the photovoltaic inverter and the energy storage battery management system into a structured JSON data stream; The data standardization encapsulation unit is used to add a device identifier and a timestamp to the JSON data stream. The device identifier includes a device type code and a geographical location code; The OPC UA server interface unit is used to publish the encapsulated data as an OPC UA node.

3. The distributed energy collaborative scheduling optimization method based on edge computing according to claim 2, characterized in that: In Step 3, the specific formulas for the adaptive attenuation mechanism to dynamically adjust the crossover rate and mutation rate of the genetic algorithm are: In formula (1), P c0 , P m0 are the initial crossover rate and mutation rate, α and β are attenuation coefficients, and t is the number of iterations.

4. The distributed energy collaborative scheduling optimization method based on edge computing according to claim 3, characterized in that: The attenuation coefficient is dynamically adjusted according to the population diversity: When the population diversity index is lower than the preset threshold, increase the attenuation coefficient to accelerate convergence and increase the coefficient γ inc ∈ [1.1, 2.0]; When the population diversity index is higher than the preset threshold, decrease the attenuation coefficient to maintain the search range and decrease the coefficient γ dec ∈ [0.5, 0.9].

5. A distributed energy collaborative scheduling optimization method based on edge computing according to claim 1, characterized in that: In Step 3, the deep reinforcement learning model adopts a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm model, which is used to optimize the initial values of the crossover rate and mutation rate of the genetic algorithm and the benchmark value of the attenuation coefficient. The state space, action space, and reward function of the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm are respectively: The state space: s t = [S f , t, ΔE](2) In formula (2), S f is the population fitness variance, t is the number of iterations, and ΔE is the energy supply-demand deviation; The action space is: a t = [ΔP c0 , ΔP m0 , Δα base , Δβ base (3) In formula (3), ΔP c0 Initial crossover rate adjustment amount, ΔP m0 Initial mutation rate adjustment amount, Δα base and Δβ base are attenuation coefficient reference value adjustment amounts; The reward function is: R = η1·(f prev - f current ) + η2·D(4) In formula (4), f prev and f current are the optimal fitness values of the previous and current generations, D is the population diversity index, and η1 and η2 are weight coefficients.

6. The distributed energy collaborative scheduling optimization method based on edge computing according to claim 5, characterized in that: The specific method for the dual-delay deep deterministic policy gradient (TD3) algorithm model to collaboratively and adaptively decay the mechanism to dynamically adjust the crossover rate and mutation rate of the genetic algorithm is as follows: S1. Before each iteration of the genetic algorithm, the dual-delay deep deterministic policy gradient (TD3) algorithm model updates the initial values and decay coefficient reference values of the crossover rate and mutation rate: In Formulas (5) and (6), t is the number of iterations, and are the updated initial values of the crossover rate and the mutation rate, and are the current initial values of the crossover rate and the mutation rate, ΔP c0 and ΔP m0 are the adjustment amounts of the crossover rate and the mutation rate, and are the updated reference values of the attenuation coefficient, and are the current reference values of the attenuation coefficient, and are the adjustment amounts of the reference value of the attenuation coefficient; S2. During the operation of the genetic algorithm, the actual decay coefficient is dynamically corrected through the population diversity index: In Formulas (7) and (8), α (t) and β (t) are adjusted attenuation coefficients, and is the attenuation coefficient reference value, D is the population diversity index, D threshold is the population diversity threshold, γ inc and γ dec are adjustment coefficients, and γ inc > 1, γ dec < 1; S3. The actual crossover rate and mutation rate are calculated according to the updated parameters through the adaptive decay mechanism: In formula (9), P c and P m are the updated crossover rate and mutation rate, and are the updated crossover rate and mutation rate, α (t) , β (t) are the updated attenuation coefficients, and t is the number of iterations.

7. A distributed energy collaborative scheduling optimization method based on edge computing according to any one of claims 1, characterized in that: In step 4, the objective function of the multi-objective particle swarm optimization algorithm is: min{w1·f loss +w2·f load +w3·f carbon}(10) In formula (10), f loss is the energy transmission loss, f load is the node load imbalance index, f carbon is the carbon emission, w1, w2, and w3 are weight coefficients, and satisfy w1 + w2 + w3 = 1.

8. A distributed energy collaborative scheduling optimization method based on edge computing according to claim 1, characterized in that: In step 5, the summary value of the global optimal scheduling policy is generated through the Merkle tree structure, and the consortium chain smart contract executes the following steps: L1. The global optimal scheduling policy is split into multiple data blocks through dynamic priorities, and priority tags are added to the headers of the data blocks; L2. According to the priority tags, the hash values of each data block are calculated hierarchically and a Merkle tree is constructed. When the scheduling policy is locally updated, only the hash values of the affected data blocks and paths are recalculated; L3. The edge computing nodes use the BLS threshold signature algorithm to generate distributed signatures. After the signature data is bound to the Merkle root hash, the summary value of the global optimal scheduling policy is written into the smart contract, triggering a deposit event; L4. The system status is monitored in real time through the smart contract rule engine, and an emergency deposit is triggered when the load suddenly changes or the battery level of the energy storage device is too low; L5. The Merkle verification paths of each data block are pre-computed and compressed and stored in the cloud control center, and each edge computing node caches the paths after the first verification.

9. The distributed energy collaborative scheduling optimization method based on edge computing according to claim 1, characterized in that: In step 5, the improved PoA consensus mechanism includes: H1. Calculate the reputation scores of each edge computing node. The formula for calculating the reputation score of the edge computing node is: In formula (11), N verify is the number of successful verifications, T response is the response time, A data is the data accuracy rate, and λ1, λ2, and λ3 are weight coefficients, and satisfy λ1 + λ2 + λ3 = 1; H2. Perform verification grading through reputation scores: When S≥80, the edge computing node is a high-reputation node and is responsible for block generation and signature; When 60≤S<80, the edge computing node is an alternative node and only participates in hash verification and has no right to generate blocks or signatures; When S<60, the edge computing node is a low-reputation node, and the verification permission is suspended and needs to be restored after manual review; H3. The smart contract generates a new list of verification nodes based on the latest reputation scores at the end of each 24-hour cycle; H4. When multiple edge computing nodes submit the same incorrect data in a short period of time, collusion detection is triggered to determine that the nodes are colluding nodes, and the integral of the colluding nodes is cleared and removed from the verification list for at least 48 hours.

10. A distributed energy collaborative scheduling optimization method based on edge computing according to claim 1, characterized in that: In step 6, the generation method of the dynamic time window historical optimal scheduling scheme includes: K1. Dynamically match the time window length according to the historical weather type, load pattern, and energy storage SOC interval; K2. Use a sliding window to calculate the correlation coefficient between the current data and historical data, and filter out historical periods with a correlation coefficient exceeding the set threshold; K3. Screen the top K optimal scheduling schemes from the historical periods by calculating the Euclidean distance, and generate an emergency plan after weighted averaging.

Images