Multi-mode coupling industrial park energy management and distribution method and system

Abstract

The invention discloses a multi-mode coupling industrial park energy management and distribution method and system, and belongs to the technical field of power system operation and control, and the method comprises the steps: collecting the operation state data of the storage, charging and discharging total factors of an internal source network load of an industrial park, and recognizing the type and real-time state of equipment through a non-intrusive load recognition algorithm; inputting a long short-term memory network to predict the photovoltaic power and the total load, outputting the maximum interruptible load capacity through fuzzy reasoning, calculating a power mutual aid reference value by taking the DC bus voltage as a constraint, solving a collaborative optimization model based on a quadratic programming algorithm, and outputting an optimal control vector to drive the adjustable equipment and the flexible interconnection converter. According to the method, the PID control parameters are adaptively adjusted according to the equipment type, the industrial load response lag is compensated, the low-voltage flexible direct-current interconnection system is used for executing power mutual aid, the power unbalance degree of the transformer area is eliminated, the voltage deviation is restrained, and collaborative optimization and accurate distribution of the source network load storage, charging and discharging total factors of the industrial park are achieved.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and control technology, specifically relating to a method and system for the management and distribution of energy in multi-mode coupled industrial parks. Background Technology

[0002] With the advancement of industrialization, industrial parks have evolved into complex energy systems integrating multiple elements such as "source, grid, load, storage, charging, and discharging." Distributed photovoltaic power, energy storage equipment, electric vehicle charging piles, and various industrial production loads are deployed within industrial parks. The management and allocation of energy in industrial parks essentially utilizes physical connections and digital technologies to monitor and schedule the flow and volume of electricity within the industrial park in real time. The management and allocation work is mainly based on the production energy needs of different enterprises, the real-time output of distributed photovoltaic power, and the peak and off-peak electricity prices of the power grid. It coordinates the consumption of photovoltaic power generation, the charging and discharging of energy storage equipment, the charging of electric vehicles, and the operation of production line loads. Through the unified allocation of resources from multiple ends, electricity is accurately delivered from the supply end to the consumption end and the storage end, realizing the orderly flow and on-demand supply of electricity and heat between different voltage levels and different power supply areas, so as to ensure the continuous satisfaction of the diverse energy needs within the industrial park.

[0003] To address the issues of uneven power distribution and voltage quality degradation caused by dual fluctuations in power sources and loads within industrial parks, existing technologies employ AC contactor switching and fixed-parameter control based on total load monitoring data. However, these methods suffer from several drawbacks: inability to smoothly and continuously adjust the mutual power supply; neglect of the inertial characteristics of inductive loads and the lag effect of frequency converter load response on control command execution; large DC bus voltage fluctuations; low accuracy of adjustable equipment tracking control commands; inability to accurately eliminate power imbalances in power distribution areas; and reduced operational stability of the industrial park's energy system. Summary of the Invention

[0004] The purpose of this invention is to provide a management and allocation method and system for energy in multi-mode coupled industrial parks. By adaptively adjusting PID control parameters according to equipment type, the lag in industrial load response is compensated. Power mutual assistance is performed using a low-voltage flexible DC interconnection system, eliminating power imbalance in transformer areas and suppressing voltage deviation. This achieves coordinated optimization and precise allocation of all elements of the industrial park, including source, grid, load, storage, charging, and discharging.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-mode coupled energy management and allocation method for industrial parks, characterized by comprising the following steps: Collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of transformer area and load regulation potential, and use non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data includes at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The preprocessed operating status data is input into a source-load trend prediction model based on a long short-term memory network to predict photovoltaic power and total load. Based on the equipment type, load regulation potential, and real-time status, the maximum interruptible load capacity at the current moment is output through fuzzy inference. With the power imbalance of the transformer area and the DC bus voltage as constraints, a transformer area mutual assistance evaluation model based on an adaptive dynamic allocation strategy is constructed, and the power mutual assistance reference value is calculated. An objective function is constructed that includes voltage deviation, total operating cost, and power mutual reference value deviation. With the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, a collaborative optimization model based on a quadratic programming algorithm is solved to output the optimal control vector, which drives the adjustable equipment and flexible interconnected converter.

[0006] Preferably, the drive adjustable device and flexible interconnect converter specifically include: Perform physical boundary exceedance verification on the optimal control vector, and send it to the underlying execution unit after the verification passes; The drive field energy controller uses a multi-objective optimization PID control algorithm to track and execute the optimal control vector. The PID control algorithm adaptively adjusts the proportional coefficient, integral coefficient, and derivative coefficient according to the identified equipment type to compensate for the response lag of different types of industrial loads. The actions performed by the underlying execution unit include controlling the on / off state of outgoing line switches, adjusting the opening and closing of intelligent circuit breakers, issuing load guidance interaction signals, performing frequency modulation on frequency converters, and controlling flexible interconnected converters. Based on the predicted photovoltaic power and the real-time status of the identified adjustable equipment, a photovoltaic power consumption analysis report, a load characteristic analysis report, and energy use optimization suggestions are generated. The digital twin interactive interface is driven to render a virtual scene of the physical entity of the industrial park based on preprocessed operating status data, dynamically displaying the predicted trends of DC bus voltage fluctuations, photovoltaic power and total load, as well as the power flow data of the flexible interconnected converter.

[0007] Preferably, the process of collecting and preprocessing operational status data of all elements within the industrial park, including source, grid, load, storage, charging, and releasing, includes: Multifunctional smart meters and power quality monitoring terminals are deployed on the low-voltage busbar side of each power supply area in the industrial park. Non-intrusive load sensing devices are connected in series at the main incoming line of the industrial park and the incoming line of each production workshop. Data acquisition gateways are deployed at the output end of the distributed photovoltaic inverter, the communication port of the battery management system of the energy storage device, and the controller port of the electric vehicle charging pile. A low-voltage flexible DC interconnection system is deployed between the AC low-voltage busbars of adjacent power supply areas. The low-voltage flexible DC interconnection system consists of flexible interconnection converters connecting each AC low-voltage busbar and DC busbars connecting the DC side of each flexible interconnection converter. Voltage transformers and current transformers are installed on the AC side and DC side of the flexible interconnection converter, respectively. Each data acquisition terminal is connected to the field energy controller through optical fiber and industrial Ethernet. Using multi-functional smart meters, non-intrusive load sensing devices, and data acquisition gateways, analog quantities and status quantities of all elements of power generation, grid, load, storage, charging, and discharging within the industrial park are collected synchronously at a preset sampling frequency. The analog quantities include the three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, high-frequency voltage and current data collected by non-intrusive load sensing equipment for each power supply area, as well as the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The power flow data includes the DC side current and AC side power of the flexible interconnection converter. State quantities include switch position signals, energy storage device state of charge, and electric vehicle connection status; The on-site energy controller performs outlier removal and time-series alignment preprocessing on the analog and state quantities of all elements of power generation, grid, load, storage, charging and discharging in the industrial park to obtain preprocessed operating status data.

[0008] Preferably, the process of identifying the device type and real-time status of the adjustable device includes: Based on the three-phase voltage and three-phase active power in the preprocessed operating status data, calculate the voltage deviation and power imbalance of the transformer area at the current moment. The voltage deviation is obtained by dividing the difference between the three-phase voltage collected at the current moment and the system rated voltage by the system rated voltage. The power imbalance of the transformer substation is obtained by dividing the difference between the maximum and minimum values ​​of the three-phase active power by the average value of the three-phase active power. The non-intrusive load identification algorithm is called to process the high-frequency voltage and current data uploaded by the non-intrusive load sensing device, construct a voltage-current trajectory image, and extract the geometric feature parameters of the voltage-current trajectory image as the load feature vector. The load feature vector is input into the pre-trained load feature library for matching to identify the equipment type and real-time status of the electrical equipment connected to the power grid at the current moment. The equipment type includes adjustable equipment and non-adjustable equipment. Adjustable equipment includes inductive load equipment and variable frequency load equipment. The real-time status includes start-stop status and operating power. Based on the identified equipment type and real-time status, combined with the rated power of adjustable equipment, the load regulation potential is calculated.

[0009] Preferably, the process of predicting photovoltaic power and total load includes: Historical photovoltaic power and historical total load are extracted from the preprocessed operating status data. Max-min normalization is performed to obtain normalized values ​​of photovoltaic power and total load. The normalized values ​​of photovoltaic power and total load are extracted using a sliding time window to construct an input feature matrix. The input feature matrix consists of input feature vectors arranged in chronological order. Each input feature vector contains the normalized values ​​of photovoltaic power and total load at the corresponding time. The input feature matrix is ​​input into the source load trend prediction model based on long short-term memory network. The source load trend prediction model uses forget gate, input gate and output gate to process each input feature vector contained in the input feature matrix in turn according to time step. When processing the input feature vector of the first time step, the hidden state vector and the cell state vector of the previous time step are set to use the preset initial zero vector. The forget gate is calculated by performing a linear weighted summation and sigmoid activation operation on the hidden state vector from the previous time step and the input feature vector from the current time step. The input gate is calculated by performing a linear weighted summation and sigmoid activation operation on the hidden state vector from the previous time step and the input feature vector from the current time step. The source load trend prediction model generates candidate cell states by performing linear weighting and hyperbolic tangent activation operations on the hidden state vector of the previous time step and the input feature vector of the current time step. The cell state vector of the previous time step is multiplied element-wise with the forget gate value, and the candidate cell state is multiplied element-wise with the input gate value. The results of the two multiplication operations are added together to obtain the cell state vector of the current time step. The cell state vector of the current time step contains the time-series characteristic information of the filtered historical photovoltaic power and historical total load. The source load trend prediction model calculates the output gate value by performing a linear weighted summation and sigmoid function activation operation on the hidden state vector of the previous time step and the input feature vector of the current time step. Perform hyperbolic tangent activation on the cell state vector at the current time step, and multiply the result element-wise with the output gate value to obtain the hidden state vector at the current time step. Input the hidden state vector at the current moment into the fully connected output layer of the source load trend prediction model; By using the weight matrix and bias vector of the fully connected output layer to perform a linear transformation on the hidden state vector, a normalized prediction result is obtained. The normalized prediction results are denormalized to output the photovoltaic power and total load for a future preset time period.

[0010] Preferably, the process of outputting the maximum interruptible load capacity at the current moment through fuzzy inference includes: Based on the identified equipment type and real-time status, select the electrical equipment that is in operation; Extract the operating power of each electrical device from the real-time status, calculate the sum of the operating power of the electrical devices in operation, and obtain the total load power at the current moment; From the electrical equipment in operation, select the set of equipment types that belong to adjustable equipment, calculate the sum of the operating power of the adjustable equipment, and obtain the adjustable load power at the current moment; The proportion of adjustable load is calculated by dividing the adjustable load power by the total load power. Establish the input and output linguistic variables of the fuzzy inference system; Set the load adjustment potential and the adjustable load ratio as input linguistic variables of the fuzzy inference system, and set the maximum interruptible load capacity as output linguistic variable of the fuzzy inference system. Establish a set of fuzzy linguistic values ​​for input and output linguistic variables. The set of fuzzy linguistic values ​​includes five fuzzy linguistic values: negative small, zero, positive small, positive medium, and positive large. Construct a membership function and use the membership function to map the clear values ​​of load adjustment potential and the clear values ​​of adjustable load ratio to the membership degrees of the corresponding fuzzy linguistic values; Construct a fuzzy rule base, which consists of fuzzy rules. Each fuzzy rule is used to define the fuzzy linguistic value that the maximum interruptible load capacity should be when the load adjustment potential and the adjustable load ratio are in a fuzzy state. For each fuzzy rule, the trigger strength is calculated using fuzzy intersection operation. The trigger strength is determined by the minimum membership degree of the two input linguistic variables in each fuzzy rule. Based on the membership function of the trigger strength and the fuzzy linguistic value corresponding to the fuzzy rule, perform fuzzy implication operation to obtain the output fuzzy set derived from the fuzzy rule; By using fuzzy union operation, the output fuzzy sets derived from each fuzzy rule in the fuzzy rule base are aggregated to obtain the total output fuzzy set; The centroid method is used to defuzzify the total output fuzzy set to obtain the maximum interruptible load capacity at the current moment.

[0011] Preferably, the process of calculating the power mutual reference value includes: Extract the DC bus voltage of the low-voltage flexible DC interconnection system from the preprocessed operating status data, obtain the power imbalance of the distribution area and the average value of the three-phase active power used when calculating the power imbalance of the distribution area, and use it to calculate the theoretical mutual assistance power requirement required to eliminate the power imbalance of the distribution area. A dynamic response gain function is constructed in the evaluation model of mutual assistance between power distribution areas based on an adaptive dynamic allocation strategy. The dynamic response gain function is used to map the power imbalance of power distribution areas into a dynamic response coefficient with a value range of zero to one. A DC voltage safety constraint function is constructed in the transformer area mutual assistance assessment model based on an adaptive dynamic allocation strategy. The DC voltage safety constraint function is used to map the DC bus voltage to a voltage safety factor with a value range of zero to one. Based on the combined theoretical power demand, dynamic response coefficient, and voltage safety factor, the power balance reference value is calculated.

[0012] Preferably, the process of constructing an objective function that includes voltage deviation, total operating cost, and power mutual reference value deviation terms includes: Define the decision variable vector for the collaborative optimization model. The decision variable vector includes the grid interaction power, energy storage charging and discharging power, electric vehicle V2G charging and discharging power, interrupted load execution power, and flexible interconnection converter mutual assistance power at the current moment. The objective function of the collaborative optimization model is constructed by a weighted sum of the voltage deviation term, the total factor operating cost term, and the mutual power reference value deviation term. The AC low-voltage busbars of each power supply area are selected as key nodes. Based on the network topology and line impedance parameters of the industrial park, the voltage-power sensitivity coefficient of each key node is calculated in advance. The voltage deviation of the key node is transformed into a quadratic function related to the decision variable vector using the voltage-power sensitivity coefficient to obtain the voltage deviation term. The total operating cost includes grid interaction cost, interruption load compensation cost, energy storage loss cost, and electric vehicle battery life loss cost. The grid interaction cost is determined by the grid time-of-use electricity price, and the interruption load compensation cost is determined by the preset interruption load compensation unit price. The energy storage loss cost and the electric vehicle battery life loss cost are modeled as quadratic functions of energy storage charging and discharging power and electric vehicle V2G charging and discharging power, respectively, to characterize the nonlinear accelerated aging effect of high-rate charging and discharging on battery life. The square of the Euclidean distance between the mutual power of the flexible interconnected converter and the power mutual reference value is calculated to obtain the deviation term of the mutual power reference value.

[0013] Preferably, the process of outputting the optimal control vector includes: The constraints for constructing the collaborative optimization model include power balance constraints, equipment physical constraints, and load regulation constraints. Establish power balance constraints, requiring that the sum of grid interaction power and predicted photovoltaic power equals the predicted total load minus the interrupted load execution power, plus the energy storage charging and discharging power, electric vehicle V2G charging and discharging power, and flexible interconnection converter mutual assistance power; Establish physical constraints on the equipment, set upper and lower limits for the charging and discharging power of energy storage, the charging and discharging power of electric vehicle V2G and the mutual assistance power of flexible interconnection converter, and limit the state of charge of energy storage devices to a preset safe range. Establish load regulation constraints, requiring that the power of interrupted loads be within the range of being greater than or equal to zero and less than or equal to the maximum interruptible load capacity; Transform the objective function and constraints into the standard mathematical form of quadratic programming. One of the interior point method and the effective set method is used to solve the co-optimization model based on the quadratic programming algorithm, and the optimal solution vector that minimizes the objective function is obtained. The optimal solution vector is output as the optimal control vector, driving the adjustable equipment and flexible interconnected converter to perform corresponding power regulation actions.

[0014] A multi-mode coupled energy management and distribution system for industrial parks, used to implement the above methods, includes: The sensing and identification module is used to collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of the distribution area and load regulation potential, and adopt a non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data includes at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The trend prediction module is used to input the preprocessed operating status data into the source-load trend prediction model based on the long short-term memory network to predict photovoltaic power and total load. The capacity assessment module is used to output the maximum interruptible load capacity at the current moment based on equipment type, load regulation potential and real-time status through fuzzy inference. The mutual assistance assessment module is used to construct a mutual assistance assessment model for transformer areas based on an adaptive dynamic allocation strategy, with the power imbalance of the transformer area and the DC bus voltage as constraints, and to calculate the power mutual assistance reference value. The collaborative optimization module is used to construct an objective function that includes voltage deviation, total operating cost, and power mutual reference value deviation. Taking the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, it solves the collaborative optimization model based on the quadratic programming algorithm and outputs the optimal control vector to drive the adjustable equipment and flexible interconnected converter.

[0015] The beneficial effects of this invention are: This invention utilizes the DC bus voltage of the low-voltage flexible DC interconnection system as a constraint to construct a transformer area mutual assistance evaluation model, calculates the power mutual assistance reference value, and drives the flexible interconnection converter to operate. This achieves smooth and continuous adjustment of mutual assistance power between transformer areas, avoids the current surge and discontinuous adjustment caused by the traditional AC contactor switching method, accurately eliminates the power imbalance of transformer areas, reduces the DC bus voltage fluctuation amplitude, and ensures the voltage quality of the low-voltage flexible DC interconnection system.

[0016] This invention decouples and identifies the equipment types of adjustable equipment in industrial parks through a non-intrusive load identification algorithm. Based on the inertial characteristics of inductive loads and the response characteristics of variable frequency loads corresponding to the equipment types, it adaptively adjusts the proportional coefficient, integral coefficient, and derivative coefficient of the PID control algorithm. This compensates for the response lag of different types of industrial loads to control commands, improves the execution accuracy and response speed of adjustable equipment to the optimal control vector, and ensures that the underlying execution unit can accurately track the upper-level optimization commands.

[0017] This invention comprehensively considers the predicted photovoltaic power and total load, the maximum interruptible load capacity, and the total operating cost. By solving a collaborative optimization model based on a quadratic programming algorithm, it generates an optimal control vector that includes the actions of flexible interconnected converters and adjustable equipment. This achieves collaborative optimization and precise allocation of all elements of energy generation, grid, load, storage, charging, and discharging within the industrial park, reduces the system operation risk caused by dual fluctuations in energy sources and loads, and improves the overall operational stability and economic benefits of the industrial park's energy system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural block diagram of the system of the present invention; Figure 3 This is a diagram showing the full-element system architecture and power flow of the industrial park's power generation, grid, load, storage, charging, and discharging system in Example 1. Figure 4 This is the logic diagram for evaluating the maximum interruptible load capacity based on fuzzy inference in Example 1; Figure 5 This is a schematic diagram of the transformer area mutual assistance evaluation model based on the adaptive dynamic allocation strategy in Example 1; Figure 6 This is the adaptive adjustment response diagram of the multi-objective optimized PID control algorithm when the device is identified as an inductive load in Example 1; Figure 7 The adaptive adjustment response diagram of the multi-objective optimized PID control algorithm when the device is identified as a frequency conversion load device in Example 1 is shown. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0020] Example 1: As Figure 1 As shown, the energy management and allocation method for multi-mode coupled industrial parks includes the following steps S1-S5: S1. Collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of transformer area and load regulation potential, and use non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data shall include at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter.

[0021] Multifunctional smart meters and power quality monitoring terminals are deployed on the low-voltage busbar side of each power supply area in the industrial park. Non-intrusive load sensing devices are connected in series at the main incoming line of the industrial park and the incoming line of each production workshop. Data acquisition gateways are deployed at the output end of the distributed photovoltaic inverter, the communication port of the battery management system of the energy storage device, and the controller port of the electric vehicle charging pile. A low-voltage flexible DC interconnection system is deployed between the AC low-voltage buses of adjacent power supply areas. The low-voltage flexible DC interconnection system consists of flexible interconnection converters connecting each AC low-voltage busbar and DC buses connecting the DC side of each flexible interconnection converter. Voltage transformers and current transformers are installed on the AC side and DC side of the flexible interconnection converter, respectively. Each data acquisition terminal is connected to the field energy controller through optical fiber and industrial Ethernet.

[0022] Using multi-functional smart meters, non-intrusive load sensing devices, and data acquisition gateways, analog and status quantities of all elements of power generation, grid, load, storage, charging, and discharging within the industrial park are collected synchronously at a preset sampling frequency.

[0023] The analog quantities include the three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, high-frequency voltage and current data collected by non-intrusive load sensing devices for each power supply area, as well as the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The power flow data includes the DC side current and AC side power of the flexible interconnection converter.

[0024] State variables include switch position signals, energy storage device charge status, and electric vehicle connection status.

[0025] The on-site energy controller performs outlier removal and time-series alignment preprocessing on the analog and state quantities of all elements of power generation, grid, load, storage, charging and discharging in the industrial park to obtain preprocessed operating status data.

[0026] Based on the three-phase voltage and three-phase active power in the preprocessed operating status data, calculate the voltage deviation and power imbalance of the transformer area at the current moment.

[0027] The voltage deviation is obtained by dividing the difference between the three-phase voltage collected at the current moment and the system rated voltage by the system rated voltage.

[0028] The power imbalance of the transformer substation is obtained by dividing the difference between the maximum and minimum values ​​of the three-phase active power by the average value of the three-phase active power.

[0029] The non-intrusive load identification algorithm is called to process the high-frequency voltage and current data uploaded by the non-intrusive load sensing device, construct a voltage-current trajectory image, and extract the geometric feature parameters of the voltage-current trajectory image as the load feature vector.

[0030] The load feature vector is input into the pre-trained load feature library for matching to identify the equipment type and real-time status of the electrical equipment connected to the power grid at the current moment. The equipment type includes adjustable equipment and non-adjustable equipment. Adjustable equipment includes inductive load equipment and variable frequency load equipment. The real-time status includes start-stop status and operating power.

[0031] Based on the identified device type and real-time status, combined with the rated power of the adjustable device Calculate load regulation potential Where N represents the total number of adjustable devices identified as being in operation. This represents the operating power of the i-th adjustable device. This represents the adjustment attribute coefficient of the i-th adjustable device. When the identified device type is a non-adjustable device... A value of 0 indicates that the identified device type is an adjustable device. The value is 1.

[0032] S2. Input the preprocessed operating status data into the source-load trend prediction model based on long short-term memory network to predict photovoltaic power and total load.

[0033] Historical photovoltaic power and historical total load are extracted from the preprocessed operating status data. Max-min normalization is performed to obtain normalized values ​​of photovoltaic power and total load. The normalized values ​​of photovoltaic power and total load are extracted using a sliding time window to construct an input feature matrix. The input feature matrix consists of input feature vectors arranged in chronological order. Each input feature vector contains the normalized values ​​of photovoltaic power and total load at the corresponding time.

[0034] The input feature matrix is ​​input into the source-load trend prediction model based on the long short-term memory network. The source-load trend prediction model uses the forget gate, input gate and output gate to process each input feature vector contained in the input feature matrix in turn according to the time step. When processing the input feature vector of the first time step, the hidden state vector and the cell state vector of the previous time step are set to use a preset initial zero vector.

[0035] The forget gate calculates its value by performing a linear weighted summation and sigmoid activation operation on the hidden state vector from the previous time step and the input feature vector from the current time step. The forget gate value determines the proportion of cell state information retained from the previous time step.

[0036] The input gate is calculated by performing a linear weighted summation and sigmoid function activation operation on the hidden state vector of the previous time step and the input feature vector of the current time step. The input gate value determines the proportion of new information written at the current time step.

[0037] The source-load trend prediction model generates candidate cell states by performing linear weighting and hyperbolic tangent activation operations on the hidden state vector of the previous time step and the input feature vector of the current time step.

[0038] The cell state vector from the previous time step is multiplied element-wise with the forget gate value, and the candidate cell state is multiplied element-wise with the input gate value. The results of the two multiplication operations are added together to obtain the cell state vector at the current time step. The cell state vector at the current time step contains the time-series characteristic information of the filtered historical photovoltaic power and historical total load.

[0039] The source-load trend prediction model calculates the output gate value by performing a linear weighted summation and sigmoid function activation operation on the hidden state vector of the previous time step and the input feature vector of the current time step.

[0040] Perform hyperbolic tangent activation on the cell state vector at the current time step, and multiply the result element-wise with the output gate value to obtain the hidden state vector at the current time step.

[0041] Input the hidden state vector at the current moment into the fully connected output layer of the source load trend prediction model.

[0042] Using the weight matrix of the fully connected output layer With bias vector For the hidden state vector Perform a linear transformation to obtain the normalized prediction result.

[0043] The normalized prediction results are denormalized, and the photovoltaic power and total load for the future preset time period are output. ,in, This represents the output vector that includes photovoltaic power and total load. This represents the vector of the maximum historical photovoltaic power and historical total load. This represents the vector of the minimum historical photovoltaic power and historical total load.

[0044] S3. Based on the equipment type, load regulation potential, and real-time status, output the maximum interruptible load capacity at the current moment through fuzzy inference.

[0045] Based on the identified device type and real-time status, select the electrical devices that are in operation.

[0046] Extract the operating power of each electrical device from the real-time status, calculate the sum of the operating power of the electrical devices in operation, and obtain the total load power at the current moment.

[0047] From the electrical equipment in operation, select the set of equipment types that are adjustable, calculate the sum of the operating power of the adjustable equipment, and obtain the adjustable load power at the current moment.

[0048] The adjustable load percentage is calculated by dividing the adjustable load power by the total load power. The adjustable load percentage reflects the proportion of responsive equipment in the load structure connected to the power grid at the current moment.

[0049] Establish the input and output linguistic variables of the fuzzy inference system.

[0050] The load adjustment potential and the adjustable load ratio are set as input linguistic variables of the fuzzy inference system, and the maximum interruptible load capacity is set as the output linguistic variable of the fuzzy inference system.

[0051] Establish a set of fuzzy linguistic values ​​for the input and output linguistic variables. The set of fuzzy linguistic values ​​includes five fuzzy linguistic values: negative small, zero, positive small, positive medium, and positive large.

[0052] Construct a membership function, and use the membership function to map the clear values ​​of load adjustment potential and the clear values ​​of the proportion of adjustable load to the membership degrees of the corresponding fuzzy linguistic values. ,in, The definite value x of the input linguistic variable represents the membership degree of one of the five fuzzy linguistic values, ranging from zero to one. c represents the Gaussian function center value corresponding to the fuzzy linguistic value. The Gaussian function width parameter represents the fuzzy language value.

[0053] A fuzzy rule base is constructed, which consists of fuzzy rules. Each fuzzy rule is used to define the fuzzy linguistic value that the maximum interruptible load capacity should be when the load adjustment potential and the adjustable load ratio are in a fuzzy state.

[0054] For each fuzzy rule, the trigger strength is calculated using fuzzy intersection operation. The trigger strength is determined by the minimum membership degree of the two input linguistic variables in each fuzzy rule, where, Indicates load regulation potential It belongs to the fuzzy set of the antecedents of the k-th rule. membership degree Indicates the adjustable load percentage It belongs to the fuzzy set of the antecedents of the k-th rule. The degree of membership.

[0055] Based on trigger strength The membership function of the fuzzy linguistic value corresponding to the k-th fuzzy rule is used to perform fuzzy implication operations to obtain the output fuzzy set derived from the k-th fuzzy rule. The specific calculation formula is as follows: ; in, Let represent the membership function of the output fuzzy set derived from the k-th fuzzy rule, and y represent the range of values ​​for the maximum interruptible load capacity. Denotes the fuzzy set of the consequent of the k-th fuzzy rule. The membership function.

[0056] By using fuzzy union operation, the output fuzzy sets derived from each fuzzy rule in the fuzzy rule base are aggregated to obtain the total output fuzzy set. The specific calculation formula is as follows: ; in, Let M represent the membership function of the total output fuzzy set, and M represent the total number of fuzzy rules in the fuzzy rule base. Let f(k) represent the membership function of the output fuzzy set derived from the k-th fuzzy rule.

[0057] The total output fuzzy set is defuzzified using the centroid method to obtain the maximum interruptible load capacity at the current moment. ,in, This represents the lower limit of the range of maximum interruptible load capacity. This represents the upper limit of the range of maximum interruptible load capacity, and y represents the integral variable, corresponding to the possible values ​​of the maximum interruptible load capacity. This represents the membership function of the total output fuzzy set.

[0058] S4. Using the power imbalance of the transformer area and the DC bus voltage as constraints, construct a transformer area mutual assistance evaluation model based on an adaptive dynamic allocation strategy, and calculate the power mutual assistance reference value.

[0059] Extract the DC bus voltage of the low-voltage flexible DC interconnection system from the preprocessed operating status data to obtain the power imbalance of the distribution area. And calculate the power imbalance of the distribution area The average value of the three-phase active power used at that time Used to calculate the power imbalance of the transformer substation. Theoretical mutual power requirement The theoretical mutual power demand is used to characterize the ideal power support required to reduce the current power imbalance of the transformer area to zero.

[0060] The dynamic response gain function is constructed in the transformer area mutual assistance assessment model based on the adaptive dynamic allocation strategy. This dynamic response gain function is used to assess the power imbalance of the transformer areas. The mapping is to dynamic response coefficients with values ​​ranging from zero to one. ,in, This represents the preset sensitivity adjustment factor. The dynamic response coefficient is used to adjust the sensitivity of the mutual assistance control. When the power imbalance of the distribution area is small, the dynamic response coefficient is small to prevent the system from adjusting frequently. When the power imbalance of the distribution area is large, the dynamic response coefficient approaches one to ensure that the system responds quickly.

[0061] Construct a DC voltage safety constraint function in the transformer substation mutual assistance assessment model based on an adaptive dynamic allocation strategy, and use the DC voltage safety constraint function to adjust the DC bus voltage. Mapped to a voltage safety factor with a value range of zero to one. ,in, This indicates the rated DC bus voltage of the low-voltage flexible DC interconnect system. This represents the preset voltage constraint bandwidth parameter, used to control the voltage safety factor. The voltage safety factor is used to limit the amount of mutual assistance power when the DC bus voltage deviates, in order to prioritize the voltage stability of the low-voltage flexible DC interconnection system.

[0062] Comprehensive theoretical mutual power demand Dynamic response coefficient and voltage safety factor Calculate the power mutual reference value The power mutual assistance reference value serves as the benchmark target for power transmission by the flexible interconnected converter in the subsequent collaborative optimization model.

[0063] S5. Construct an objective function that includes voltage deviation, total operating cost, and power mutual reference value deviation. Using the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, solve the collaborative optimization model based on the quadratic programming algorithm, output the optimal control vector, and drive the adjustable equipment and flexible interconnected converter.

[0064] Define the decision variable vector of the collaborative optimization model The decision variable vector contains the power grid interaction power at the current moment. Energy storage charging and discharging power V2G charging and discharging power of electric vehicles Interruption load execution power and flexible interconnect converter mutual power Among them, grid interaction power Positive values ​​represent electricity purchased from the grid, while negative values ​​represent electricity sold to the grid; energy storage charging and discharging power. V2G charging and discharging power of electric vehicles Positive values ​​represent charging, and negative values ​​represent discharging.

[0065] Objective function for constructing a collaborative optimization model The objective function consists of the voltage deviation term. Total factor operating cost item and mutual assistance power reference value deviation item The weighted summation is used to construct the equation, where J represents the total scalar value of the objective function. These are the voltage deviation terms. Total factor operating cost item and mutual assistance power reference value deviation item The weighting coefficients, The unit is yuan per volt square hour. Dimensionless The unit is yuan per kilowatt-hour.

[0066] The AC low-voltage busbars of each power supply area are selected as key nodes. Based on the network topology and line impedance parameters of the industrial park, the voltage-power sensitivity coefficient of each key node is pre-calculated. Using the voltage-power sensitivity coefficient, the voltage deviation of the key nodes is transformed into a quadratic function related to the decision variable vector X, thus obtaining the voltage deviation term. ,in, This represents the row vector of voltage-power sensitivity coefficients for the j-th critical node. This represents the target allowable voltage deviation value for the j-th critical node.

[0067] Total factor operating cost item Includes grid interaction costs Interruption load compensation cost Energy storage loss cost and the cost of electric vehicle battery lifespan degradation Grid interaction cost Time-of-use pricing of the power grid Decision on interruption load compensation costs The pre-set interruption load compensation unit price The decision was made to include energy storage loss costs. and the cost of electric vehicle battery lifespan degradation They are modeled as energy storage charging and discharging power respectively. and V2G charging and discharging power of electric vehicles The quadratic function is used to characterize the nonlinear accelerated aging effect of high-rate charge-discharge on battery life, where, and These represent the lifespan degradation coefficients of energy storage batteries and electric vehicle batteries, respectively, expressed in yuan per kilowatt-hour.

[0068] Calculate the mutual power of flexible interconnected converters Power mutual reference value The square of the Euclidean distance between them yields the deviation term of the mutual power reference value. .

[0069] The constraints for constructing the collaborative optimization model include power balance constraints, equipment physical constraints, and load regulation constraints.

[0070] Establish power balance constraints Requires grid interaction power With the predicted photovoltaic power The sum equals the predicted total load. Subtract interrupted load execution power In addition, energy storage charging and discharging power V2G charging and discharging power of electric vehicles and flexible interconnect converter mutual power .

[0071] Establish physical constraints for the equipment and set the energy storage charging and discharging power. V2G charging and discharging power of electric vehicles and flexible interconnect converter mutual power The upper and lower limits are set, and the state of charge of the energy storage device is limited to a preset safe range.

[0072] Establish load regulation constraints Requires interruption of load execution power It is greater than or equal to zero and less than or equal to the maximum interruptible load capacity. Within the range.

[0073] The objective function and constraints are converted into the standard mathematical form of quadratic programming.

[0074] One of the interior point method and the effective set method is used to solve the co-optimization model based on the quadratic programming algorithm, and the optimal solution vector that minimizes the objective function is obtained.

[0075] The optimal solution vector is output as the optimal control vector, driving the adjustable equipment and flexible interconnected converter to perform corresponding power regulation actions.

[0076] The optimal control vector is subjected to physical boundary exceedance verification, and after the verification is passed, it is sent to the underlying execution unit.

[0077] The drive field energy controller uses a multi-objective optimization PID control algorithm to track and execute the optimal control vector. The PID control algorithm adaptively adjusts the proportional coefficient, integral coefficient, and derivative coefficient according to the identified equipment type to compensate for the response lag of different types of industrial loads.

[0078] The actions performed by the underlying execution unit include controlling the on / off state of outgoing line switches, adjusting the opening and closing of intelligent circuit breakers, issuing load guidance interaction signals, performing frequency modulation on frequency converters, and controlling flexible interconnected converters.

[0079] By combining the predicted photovoltaic power and the real-time status of the identified adjustable equipment, a photovoltaic absorption analysis report, a load characteristic analysis report, and energy use optimization suggestions are generated.

[0080] The digital twin interactive interface is driven to render a virtual scene of the physical entity of the industrial park based on preprocessed operating status data, dynamically displaying the predicted trends of DC bus voltage fluctuations, photovoltaic power and total load, as well as the power flow data of the flexible interconnected converter.

[0081] like Figures 3 to 7 As shown, through simulation experiments, the method of this embodiment is verified to achieve the ability of coordinated optimization control of all elements of source, grid, load, storage, charging and discharging and improve the stability of system operation in a mixed energy consumption scenario that includes inductive load equipment and frequency conversion load equipment.

[0082] Using an industrial park comprising two AC power supply areas and a low-voltage flexible DC interconnection system as the simulation object, and taking the operational status data of all elements of the industrial park's power generation, grid, load, energy storage, and charging / discharging within a 24-hour operating cycle as input, the simulation test diagram is shown below. Figure 3-7 As shown.

[0083] Figure 3 The diagram shows the full-element system architecture demonstration stage of the method in this embodiment. It clearly demonstrates the topology structure that connects the AC low-voltage bus of transformer area 1 and the AC low-voltage bus of transformer area 2 through a low-voltage flexible DC interconnection system (DC bus), as well as the access locations of all elements of equipment such as distributed photovoltaic, energy storage devices, and V2G charging piles. The arrows in the figure dynamically indicate the bidirectional flow and mutual assistance process of power between the AC and DC systems, verifying that the physical architecture constructed by the method in this embodiment can support cross-transformer area power mutual assistance and multi-source collaboration.

[0084] Figure 4The diagram illustrates the maximum interruptible load capacity assessment stage, demonstrating the nonlinear relationship between load adjustment potential and adjustable load ratio mapped to maximum interruptible load capacity through fuzzy inference logic. When both load adjustment potential and adjustable load ratio are high, the output maximum interruptible load capacity increases significantly. Conversely, when one of these indicators is low, the output maximum interruptible load capacity is suppressed. This nonlinear mapping relationship aligns with actual scheduling requirements, avoids biases caused by single-indicator assessments, and verifies that the method in this embodiment can intelligently determine demand response capabilities based on the real-time status of the equipment.

[0085] Figure 5 The diagram shows the power mutual assistance reference value calculation stage, displaying the adaptive curves of the dynamic response coefficient (dashed line) and voltage safety factor (solid line) as a function of simulation time (step / 15min). As the simulation time progresses, the dynamic response coefficient adjusts in real time to follow the fluctuations in the power imbalance of the transformer area. When the power imbalance of the transformer area is large, the dynamic response coefficient approaches 1 to improve the response speed. The voltage safety factor remains around 1 for most of the time. However, during the simulation process, when the DC bus voltage deviates from the rated voltage of 750V, the voltage safety factor drops rapidly to limit the mutual assistance power. This result verifies that the method of this embodiment can ensure the voltage safety and control sensitivity of the low-voltage flexible DC interconnection system by adaptively adjusting the dynamic response coefficient and voltage safety factor, based on a comprehensive consideration of the theoretical mutual assistance power requirements.

[0086] Figure 6 The diagram shows the situation when the device is identified as an inductive load during the execution control phase. The response curve shows a slow ramp with large inertia and a long rise time, which verifies that the method in this embodiment effectively compensates for the large inertia lag of the inductive load device by adaptively adjusting the proportional coefficient, integral coefficient and derivative coefficient of the PID control algorithm.

[0087] Figure 7 The figure shows the situation when the device is identified as a frequency converter load during the execution control phase. The response curve shows a rapid rise, which can quickly track the target command. This verifies that the method in this embodiment has optimized the parameters for the fast response characteristics of the frequency converter load device, ensuring the sensitivity of the control.

[0088] This simulation example verifies that the method in this example, by combining a non-intrusive load identification algorithm with a multi-objective optimization PID control algorithm, effectively solves the response lag problem caused by the difficulty of traditional fixed parameter control in adapting to the diversity of industrial loads. By introducing a low-voltage flexible DC interconnection system and a voltage safety constraint mechanism, the power balance of the distribution area and the stability of the DC bus voltage are guaranteed.

[0089] In summary, the method in this embodiment demonstrates precise sensing, intelligent decision-making, and adaptive collaborative control capabilities under dual source-load fluctuations and multiple types of load access. At the same time, through physical-level flexible DC interconnection and information-level digital twin interaction, it constructs a closed-loop management system integrating source, grid, load, storage, charging, and discharging, which has significant value for promotion and application.

[0090] Example 2: Figure 2 As shown, the multi-mode coupled industrial park energy management and distribution system is used to implement the method in Example 1, including a sensing and identification module, a trend prediction module, a capacity assessment module, a mutual assistance assessment module, and a collaborative optimization module, wherein the modules are electrically connected.

[0091] The sensing and identification module is used to collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of transformer area and load regulation potential, and adopt a non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data includes at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter.

[0092] The trend prediction module is used to input the preprocessed operating status data into the source-load trend prediction model based on long short-term memory network to predict photovoltaic power and total load.

[0093] The capacity assessment module is used to output the maximum interruptible load capacity at the current moment through fuzzy inference based on equipment type, load regulation potential, and real-time status.

[0094] The mutual assistance assessment module is used to construct a mutual assistance assessment model for transformer substations based on an adaptive dynamic allocation strategy, with the power imbalance of the substation and the DC bus voltage as constraints, and to calculate the power mutual assistance reference value.

[0095] The collaborative optimization module is used to construct an objective function that includes voltage deviation, total operating cost, and power mutual reference value deviation. With the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, it solves the collaborative optimization model based on the quadratic programming algorithm, outputs the optimal control vector, and drives the adjustable equipment and flexible interconnected converter.

[0096] The explanations and effects of the method in Example 1 are also applicable to the system in this example, and the principle is the same. Therefore, this example will not be limited further.

Claims

1. A multi-mode coupled energy management and allocation method for industrial parks, characterized in that, Includes the following steps: Collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of transformer area and load regulation potential, and use non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data includes at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The preprocessed operating status data is input into a source-load trend prediction model based on a long short-term memory network to predict photovoltaic power and total load. Based on the equipment type, load regulation potential, and real-time status, the maximum interruptible load capacity at the current moment is output through fuzzy inference. With the power imbalance of the transformer area and the DC bus voltage as constraints, a transformer area mutual assistance evaluation model based on an adaptive dynamic allocation strategy is constructed, and the power mutual assistance reference value is calculated. An objective function is constructed that includes voltage deviation, total operating cost, and power mutual reference value deviation. With the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, a collaborative optimization model based on a quadratic programming algorithm is solved to output the optimal control vector, which drives the adjustable equipment and flexible interconnected converter.

2. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 1, characterized in that, Drive adjustable devices and flexible interconnected converters, specifically including: Perform physical boundary exceedance verification on the optimal control vector, and send it to the underlying execution unit after the verification passes; The drive field energy controller uses a multi-objective optimization PID control algorithm to track and execute the optimal control vector. The PID control algorithm adaptively adjusts the proportional coefficient, integral coefficient, and derivative coefficient according to the identified equipment type to compensate for the response lag of different types of industrial loads. The actions performed by the underlying execution unit include controlling the on / off state of outgoing line switches, adjusting the opening and closing of intelligent circuit breakers, issuing load guidance interaction signals, performing frequency modulation on frequency converters, and controlling flexible interconnected converters. Based on the predicted photovoltaic power and the real-time status of the identified adjustable equipment, a photovoltaic power consumption analysis report, a load characteristic analysis report, and energy use optimization suggestions are generated. The digital twin interactive interface is driven to render a virtual scene of the physical entity of the industrial park based on preprocessed operating status data, dynamically displaying the predicted trends of DC bus voltage fluctuations, photovoltaic power and total load, as well as the power flow data of the flexible interconnected converter.

3. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 1, characterized in that, The process of collecting and preprocessing operational status data of all elements within an industrial park, including power generation, grid, load, storage, charging, and releasing, includes: Multifunctional smart meters and power quality monitoring terminals are deployed on the low-voltage busbar side of each power supply area in the industrial park. Non-intrusive load sensing devices are connected in series at the main incoming line of the industrial park and the incoming line of each production workshop. Data acquisition gateways are deployed at the output end of the distributed photovoltaic inverter, the communication port of the battery management system of the energy storage device, and the controller port of the electric vehicle charging pile. A low-voltage flexible DC interconnection system is deployed between the AC low-voltage busbars of adjacent power supply areas. The low-voltage flexible DC interconnection system consists of flexible interconnection converters connecting each AC low-voltage busbar and DC busbars connecting the DC side of each flexible interconnection converter. Voltage transformers and current transformers are installed on the AC side and DC side of the flexible interconnection converter, respectively. Each data acquisition terminal is connected to the field energy controller through optical fiber and industrial Ethernet. Using multi-functional smart meters, non-intrusive load sensing devices, and data acquisition gateways, analog quantities and status quantities of all elements of power generation, grid, load, storage, charging, and discharging within the industrial park are collected synchronously at a preset sampling frequency. The analog quantities include the three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, high-frequency voltage and current data collected by non-intrusive load sensing equipment for each power supply area, as well as the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The power flow data includes the DC side current and AC side power of the flexible interconnection converter. State quantities include switch position signals, energy storage device state of charge, and electric vehicle connection status; The on-site energy controller performs outlier removal and time-series alignment preprocessing on the analog and state quantities of all elements of power generation, grid, load, storage, charging and discharging in the industrial park to obtain preprocessed operating status data.

4. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 3, characterized in that, The process of identifying the device type and real-time status of adjustable devices includes: Based on the three-phase voltage and three-phase active power in the preprocessed operating status data, calculate the voltage deviation and power imbalance of the transformer area at the current moment. The voltage deviation is obtained by dividing the difference between the three-phase voltage collected at the current moment and the system rated voltage by the system rated voltage. The power imbalance of the transformer substation is obtained by dividing the difference between the maximum and minimum values ​​of the three-phase active power by the average value of the three-phase active power. The non-intrusive load identification algorithm is called to process the high-frequency voltage and current data uploaded by the non-intrusive load sensing device, construct a voltage-current trajectory image, and extract the geometric feature parameters of the voltage-current trajectory image as the load feature vector. The load feature vector is input into the pre-trained load feature library for matching to identify the equipment type and real-time status of the electrical equipment connected to the power grid at the current moment. The equipment type includes adjustable equipment and non-adjustable equipment. Adjustable equipment includes inductive load equipment and variable frequency load equipment. The real-time status includes start-stop status and operating power. Based on the identified equipment type and real-time status, combined with the rated power of adjustable equipment, the load regulation potential is calculated.

5. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 1, characterized in that, The process of predicting photovoltaic power and total load includes: Historical photovoltaic power and historical total load are extracted from the preprocessed operating status data. Max-min normalization is performed to obtain normalized values ​​of photovoltaic power and total load. The normalized values ​​of photovoltaic power and total load are extracted using a sliding time window to construct an input feature matrix. The input feature matrix consists of input feature vectors arranged in chronological order. Each input feature vector contains the normalized values ​​of photovoltaic power and total load at the corresponding time. The input feature matrix is ​​input into the source load trend prediction model based on long short-term memory network. The source load trend prediction model uses forget gate, input gate and output gate to process each input feature vector contained in the input feature matrix in turn according to time step. When processing the input feature vector of the first time step, the hidden state vector and the cell state vector of the previous time step are set to use the preset initial zero vector. The forget gate is calculated by performing a linear weighted summation and sigmoid activation operation on the hidden state vector from the previous time step and the input feature vector from the current time step. The input gate is calculated by performing a linear weighted summation and sigmoid activation operation on the hidden state vector from the previous time step and the input feature vector from the current time step. The source load trend prediction model generates candidate cell states by performing linear weighting and hyperbolic tangent activation operations on the hidden state vector of the previous time step and the input feature vector of the current time step. The cell state vector of the previous time step is multiplied element-wise with the forget gate value, and the candidate cell state is multiplied element-wise with the input gate value. The results of the two multiplication operations are added together to obtain the cell state vector of the current time step. The cell state vector of the current time step contains the time-series characteristic information of the filtered historical photovoltaic power and historical total load. The source load trend prediction model calculates the output gate value by performing a linear weighted summation and sigmoid function activation operation on the hidden state vector of the previous time step and the input feature vector of the current time step. Perform hyperbolic tangent activation on the cell state vector at the current time step, and multiply the result element-wise with the output gate value to obtain the hidden state vector at the current time step. Input the hidden state vector at the current moment into the fully connected output layer of the source load trend prediction model; By using the weight matrix and bias vector of the fully connected output layer to perform a linear transformation on the hidden state vector, a normalized prediction result is obtained. The normalized prediction results are denormalized to output the photovoltaic power and total load for a future preset time period.

6. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 4, characterized in that, The process of outputting the maximum interruptible load capacity at the current moment through fuzzy inference includes: Based on the identified equipment type and real-time status, select the electrical equipment that is in operation; Extract the operating power of each electrical device from the real-time status, calculate the sum of the operating power of the electrical devices in operation, and obtain the total load power at the current moment; From the electrical equipment in operation, select the set of equipment types that belong to adjustable equipment, calculate the sum of the operating power of the adjustable equipment, and obtain the adjustable load power at the current moment; The proportion of adjustable load is calculated by dividing the adjustable load power by the total load power. Establish the input and output linguistic variables of the fuzzy inference system; Set the load adjustment potential and the adjustable load ratio as input linguistic variables of the fuzzy inference system, and set the maximum interruptible load capacity as output linguistic variable of the fuzzy inference system. Establish a set of fuzzy linguistic values ​​for input and output linguistic variables. The set of fuzzy linguistic values ​​includes five fuzzy linguistic values: negative small, zero, positive small, positive medium, and positive large. Construct a membership function and use the membership function to map the clear values ​​of load adjustment potential and the clear values ​​of adjustable load ratio to the membership degrees of the corresponding fuzzy linguistic values; Construct a fuzzy rule base, which consists of fuzzy rules. Each fuzzy rule is used to define the fuzzy linguistic value that the maximum interruptible load capacity should be when the load adjustment potential and the adjustable load ratio are in a fuzzy state. For each fuzzy rule, the trigger strength is calculated using fuzzy intersection operation. The trigger strength is determined by the minimum membership degree of the two input linguistic variables in each fuzzy rule. Based on the membership function of the trigger strength and the fuzzy linguistic value corresponding to the fuzzy rule, perform fuzzy implication operation to obtain the output fuzzy set derived from the fuzzy rule; By using fuzzy union operation, the output fuzzy sets derived from each fuzzy rule in the fuzzy rule base are aggregated to obtain the total output fuzzy set; The centroid method is used to defuzzify the total output fuzzy set to obtain the maximum interruptible load capacity at the current moment.

7. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 4, characterized in that, The process of calculating the power mutual reference value includes: Extract the DC bus voltage of the low-voltage flexible DC interconnection system from the preprocessed operating status data, obtain the power imbalance of the distribution area and the average value of the three-phase active power used when calculating the power imbalance of the distribution area, and use it to calculate the theoretical mutual assistance power requirement required to eliminate the power imbalance of the distribution area. A dynamic response gain function is constructed in the evaluation model of mutual assistance between power distribution areas based on an adaptive dynamic allocation strategy. The dynamic response gain function is used to map the power imbalance of power distribution areas into a dynamic response coefficient with a value range of zero to one. A DC voltage safety constraint function is constructed in the transformer area mutual assistance assessment model based on an adaptive dynamic allocation strategy. The DC voltage safety constraint function is used to map the DC bus voltage to a voltage safety factor with a value range of zero to one. Based on the combined theoretical power demand, dynamic response coefficient, and voltage safety factor, the power balance reference value is calculated.

8. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 7, characterized in that, The process of constructing an objective function that includes terms for voltage deviation, total operating cost, and power mutual reference value deviation includes: Define the decision variable vector for the collaborative optimization model. The decision variable vector includes the grid interaction power, energy storage charging and discharging power, electric vehicle V2G charging and discharging power, interrupted load execution power, and flexible interconnection converter mutual assistance power at the current moment. The objective function of the collaborative optimization model is constructed by a weighted sum of the voltage deviation term, the total factor operating cost term, and the mutual power reference value deviation term. The AC low-voltage busbars of each power supply area are selected as key nodes. Based on the network topology and line impedance parameters of the industrial park, the voltage-power sensitivity coefficient of each key node is calculated in advance. The voltage deviation of the key node is transformed into a quadratic function related to the decision variable vector using the voltage-power sensitivity coefficient to obtain the voltage deviation term. The total operating cost includes grid interaction cost, interruption load compensation cost, energy storage loss cost, and electric vehicle battery life loss cost. The grid interaction cost is determined by the grid time-of-use electricity price, and the interruption load compensation cost is determined by the preset interruption load compensation unit price. The energy storage loss cost and the electric vehicle battery life loss cost are modeled as quadratic functions of energy storage charging and discharging power and electric vehicle V2G charging and discharging power, respectively, to characterize the nonlinear accelerated aging effect of high-rate charging and discharging on battery life. The square of the Euclidean distance between the mutual power of the flexible interconnected converter and the power mutual reference value is calculated to obtain the deviation term of the mutual power reference value.

9. The method for managing and allocating energy in a multi-mode coupled industrial park according to claim 8, characterized in that, The process of outputting the optimal control vector includes: The constraints for constructing the collaborative optimization model include power balance constraints, equipment physical constraints, and load regulation constraints. Establish power balance constraints, requiring that the sum of grid interaction power and predicted photovoltaic power equals the predicted total load minus the interrupted load execution power, plus the energy storage charging and discharging power, electric vehicle V2G charging and discharging power, and flexible interconnection converter mutual assistance power; Establish physical constraints on the equipment, set upper and lower limits for the charging and discharging power of energy storage, the charging and discharging power of electric vehicle V2G and the mutual assistance power of flexible interconnection converter, and limit the state of charge of energy storage devices to a preset safe range. Establish load regulation constraints, requiring that the power of interrupted loads be within the range of being greater than or equal to zero and less than or equal to the maximum interruptible load capacity; Transform the objective function and constraints into the standard mathematical form of quadratic programming. One of the interior point method and the effective set method is used to solve the co-optimization model based on the quadratic programming algorithm, and the optimal solution vector that minimizes the objective function is obtained. The optimal solution vector is output as the optimal control vector, driving the adjustable equipment and flexible interconnected converter to perform corresponding power regulation actions.

10. A multi-mode coupled energy management and distribution system for industrial parks, used to implement the method described in any one of claims 1-9, characterized in that, include: The sensing and identification module is used to collect and preprocess the operating status data of all elements of power generation, grid, load, storage, charging and discharging in the industrial park, obtain voltage deviation, power imbalance of the distribution area and load regulation potential, and adopt a non-intrusive load identification algorithm to identify the equipment type and real-time status of adjustable equipment. The operating status data includes at least the DC bus voltage of the low-voltage flexible DC interconnection system and the power flow data of the flexible interconnection converter. The trend prediction module is used to input the preprocessed operating status data into the source-load trend prediction model based on the long short-term memory network to predict photovoltaic power and total load. The capacity assessment module is used to output the maximum interruptible load capacity at the current moment based on equipment type, load regulation potential and real-time status through fuzzy inference. The mutual assistance assessment module is used to construct a mutual assistance assessment model for transformer areas based on an adaptive dynamic allocation strategy, with the power imbalance of the transformer area and the DC bus voltage as constraints, and to calculate the power mutual assistance reference value. The collaborative optimization module is used to construct an objective function that includes voltage deviation, total operating cost, and power mutual reference value deviation. Taking the predicted photovoltaic power and total load as inputs and the maximum interruptible load capacity as constraints, it solves the collaborative optimization model based on the quadratic programming algorithm and outputs the optimal control vector to drive the adjustable equipment and flexible interconnected converter.

Images