DC power supply coordination control system based on industrial big data

Abstract

The invention relates to the technical field of intelligent power grids and industrial automation control, in particular to a DC power supply coordination control system based on industrial big data. The system comprises a multi-dimensional heterogeneous data sensing module, a feature mapping module, a multi-target collaborative decision-making module and a self-adaptive execution module. The system constructs a load characteristic and electric energy quality correlation model based on the operation state and the power consumption demand, outputs dynamic electrical characteristics and sensitivity characteristics, and solves a power distribution instruction by taking power grid quality, load balance and efficiency as cooperative targets; according to the method, a perception decision-making framework crossing an electrical domain and a physical domain is constructed, hidden physical layer states such as impedance drift are mined through deep learning, the problem that a load physical state is ignored and production quality is disjointed in traditional control is solved, and the industrial yield and energy efficiency are improved while the quality of a power grid is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of smart grid and industrial automation control technology, specifically to a DC power supply coordination control system based on industrial big data. Background Technology

[0002] In the application scenario of DC power supply coordination control based on industrial big data, the power system relies on accurate operating status perception and power regulation strategy to ensure the stability of the distribution network and the power demand of the load side. The control system usually needs to combine the electrical parameters of the grid side and the operating condition data of the load side to adjust the output power and operating point of the power supply in real time. For the operation and control of DC power supplies, existing solutions generally adopt a single electrical index feedback architecture. This involves collecting electrical quantities at the power supply output port through voltage transformers or current sensors, and using proportional-integral-derivative algorithms or conventional feedback control logic to directly adjust the power supply's duty cycle or output commands based on voltage or current deviations. While this approach is feasible in environments with constant load characteristics or low power quality requirements, its over-reliance on a single data source in the electrical domain and neglect of the deep correlation between the physical state of the load side and the final industrial production quality makes it prone to overlooking physical factors such as load impedance drift and thermal inertia hysteresis when encountering complex and ever-changing industrial production environments. Layer changes lead to a disconnect between power supply regulation and actual process requirements. In addition, existing control systems struggle to effectively handle the spatiotemporal alignment challenges between high-frequency electrical quantity sampling and low-frequency process quantity sampling, and lack trend prediction and proactive defense capabilities based on historical data. This results in only delayed, passive corrections during grid fluctuations or load abrupt changes, making it difficult to support the high-precision industrial manufacturing requirements for power supply system coordination, high-response speed, and production line-level process quality assurance. Therefore, establishing a control mechanism with cross-domain data fusion and multi-objective collaborative decision-making capabilities to effectively eliminate data barriers while improving the DC power supply's perception accuracy of load physical characteristics and the overall system energy efficiency has become an urgent technical problem to be solved. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of this invention is to provide a DC power supply coordination control system based on industrial big data, which can solve the technical problems existing in the prior art. Specifically, the technical solution of this invention includes: In the cloud, the cloud communication connection includes a multi-dimensional heterogeneous data perception module, a feature mapping module, a multi-objective collaborative decision-making module, and an adaptive execution module; The multidimensional heterogeneous data sensing module is used to collect operating status data on the distribution network side and electricity demand data on the load side. The feature mapping module is used to construct a load characteristic and power quality correlation model based on the operating status data and power demand data, and output the dynamic electrical characteristics and sensitivity characteristics of the load. The multi-objective collaborative decision-making module is used to input the dynamic electrical characteristics and sensitivity features into a preset system coordination and optimization unit, and calculate the optimal operating point and power allocation command of the DC power supply with grid quality, load balance and system efficiency as collaborative objectives. The adaptive execution module is used to generate network coordination control instructions based on the optimized operating point and power allocation instructions, and send them to the power distribution management unit or load switch connected to the DC power supply, so as to adjust the output power of the DC power supply in the power distribution network and its coordination relationship with other power supplies.

[0004] Preferably, the process by which the multidimensional heterogeneous data sensing module collects operating status data from the distribution network side and electricity demand data from the load side includes: Acquire the voltage, frequency, and harmonic distortion rate data of the distribution network common connection point, as well as the output current, power factor, and available capacity data of the DC power supply body, mark them as operating status data, and set the millisecond-level acquisition frequency; Acquire the planned production curve, real-time active / reactive power demand, and voltage tolerance range data of key loads on the load side, mark them as power demand data, and set the second-level acquisition frequency; The operational status data and electricity demand data collected at the same time are time-series aligned to generate a multidimensional heterogeneous dataset.

[0005] Preferably, the process by which the feature mapping module constructs the correlation model between load characteristics and power quality includes: Historical multidimensional heterogeneous datasets are obtained as training samples. The trend of electricity demand data change in the training samples is extracted as input features. The correlation between the load-end voltage deviation and the fluctuation of key process parameters in the corresponding training samples is extracted as output labels. An initial correlation model is constructed using a long short-term memory network algorithm. The initial correlation model is then iteratively trained using training samples to obtain a completed correlation model between load characteristics and power quality. The real-time collected electricity demand data is input into the load characteristics and power quality correlation model, and the dynamic electrical characteristics and sensitivity features of the load under the current operating conditions are output.

[0006] Preferably, the process by which the multi-objective collaborative decision-making module calculates the optimal operating point and power allocation command for the DC power supply includes: A system coordination and optimization unit is constructed, which includes distribution network operation constraints, multi-DC power source coordination rules, and load priority strategies. Extract the real-time operating conditions of the power grid and the DC power supply status from the operating status data, and input them into the system coordination and optimization unit in combination with the dynamic electrical characteristics and sensitivity features of the load. The system coordination and optimization unit performs optimization calculations with the goal of minimizing the voltage and frequency deviation at the point of common coupling, minimizing the variance of the load rate of each power supply, and prioritizing the power supply to critical loads. It outputs the optimized operating point of the DC power supply, including the target output power, power factor, and power allocation instructions.

[0007] Preferably, the multi-objective collaborative decision-making module is further configured to perform multi-source collaborative and load scheduling operations, which include: Preset power grid quality safety range; Obtain real-time voltage or frequency data of the current point of common coupling, and predict the voltage or frequency value at the next moment based on the rate of change; Determine whether the real-time data and the predicted voltage or frequency value for the next moment are within the safe range; If the real-time data and the predicted voltage or frequency value for the next moment do not deviate from the safe range, then the current output command of each DC power supply remains unchanged. If the real-time data is within the safe range but the predicted voltage or frequency value for the next moment deviates from the safe range, it is determined that a deviation is imminent. The collaborative scheduling program is then initiated to recalculate the output combination of each controllable unit in the system, including the DC power supply and other adjustable distributed power sources / energy storage, and to generate new optimized operating points and power allocation instructions. These instructions are simultaneously sent to multiple controlled units.

[0008] Preferably, the system includes multiple DC power sources connected to the same distribution network segment, and the multi-objective collaborative decision-making module is also used to perform network power flow optimization operations, which include: Obtain the topology of the power distribution network segment, the impedance parameters of each line, and the demand forecast of all load nodes; Based on the aforementioned load dynamic electrical characteristics and sensitivity features, predict the network power flow distribution and node voltage levels under different power output schemes; With the goal of minimizing network losses and balancing line load rates, the optimal power flow is solved to obtain the optimal operating point and power allocation command for each DC power source.

[0009] Preferably, the process by which the adaptive execution module generates network coordination control instructions based on the optimized operating point and power allocation instructions includes: Obtain the current upper-level scheduling command or local setting value received by the DC power supply; Calculate the absolute value of the deviation between the optimized operating point and the power allocation command and the current command; Determine whether the absolute value of the deviation is greater than a preset adjustment dead zone threshold; If the deviation exceeds the aforementioned dead zone threshold, a control command sequence is generated based on the deviation. The control commands are issued through the power monitoring network or the energy management system communication protocol. The control objective is to adjust the output power setpoint of the DC power supply or to operate the associated grid-connected switch and power regulation unit.

[0010] Preferably, the adaptive execution module further includes a predictive power quality management unit, which performs the following operations: Based on historical and real-time operational data, the voltage sag, harmonic distortion rate increment, or power shortage that may occur in the distribution network within a preset time period can be predicted. If a predicted event will affect a critical load and the voltage sag, harmonic distortion rate increment, or power shortage exceeds a preset severity threshold, a set of preventative control instructions will be generated in advance. In response to this instruction set, the response strategies of the DC power supply, energy storage system, and reactive power compensation device are coordinated to adjust the network operating status before an event occurs, thereby achieving proactive defense.

[0011] Compared with the prior art, the present invention has the following improvements and advantages: 1. This invention effectively solves the problem that existing DC power supply control focuses only on electrical indicators and ignores the disconnect between the physical state of the load side and the final industrial production quality by constructing a perception and decision-making architecture that spans the electrical and physical domains. Unlike traditional single voltage or current feedback control, this solution uses deep learning algorithms to mine the hidden physical layer and construct a model that correlates load characteristics with power quality. It can identify load physical state changes that cannot be perceived by simple electrical control, such as impedance drift or thermal inertia hysteresis, thereby maximizing the production yield of industrial products and system energy efficiency while ensuring power grid quality. 2. This invention employs a layered acquisition and time-series alignment mechanism, solving the problem of aligning multi-physics data in the spatiotemporal dimension. By performing timestamp anchoring and interpolation calculations on the high-frequency acquisition of electrical quantities on the distribution network side and the low-frequency acquisition of process quantities on the load side, the fast variables of electrical quantities and the slow variables of process quantities are integrated on the same time axis, providing an accurate data foundation for exploring the lagging impact of voltage fluctuations on product quality, avoiding misjudgments of causal relationships caused by data asynchrony, and significantly improving the fidelity of multi-source data fusion. 3. This invention introduces multi-objective collaborative decision-making and network power flow optimization technology, realizing a leap from device-level local optimization to system-level global optimization. By coupling the dynamic electrical characteristics of the load into the network power flow constraints, and optimizing with the objectives of minimizing network loss, balancing line load rate and load rate variance, this method not only solves the circulating current problem when multiple power supplies are connected in parallel, but also prevents some power supplies from being overloaded while others are lightly loaded, effectively extending the average service life of the power supply system and improving the overall transmission efficiency. 4. This invention establishes a power quality management mechanism based on rate of change prediction and active defense, giving the system the ability to prevent problems in advance. The system uses a sequence prediction network to predict potential voltage dips, harmonic distortions, or power shortages in the future, and generates a set of preventive control instructions in advance before the events occur. It coordinates the DC power supply, energy storage system, and reactive power compensation device to respond, changing the control mode from traditional lag correction to active defense, effectively avoiding production interruptions or scrap caused by adjustment lag in precision industrial loads. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 A DC power supply coordination control system based on industrial big data includes: In the cloud, cloud communication connections include a multi-dimensional heterogeneous data perception module, a feature mapping module, a multi-objective collaborative decision-making module, and an adaptive execution module; The multidimensional heterogeneous data sensing module is used to collect operating status data on the distribution network side and electricity demand data on the load side. The feature mapping module is used to construct a correlation model between load characteristics and power quality based on operating status data and power demand data, and output the dynamic electrical characteristics and sensitivity features of the load. The multi-objective collaborative decision-making module is used to input dynamic electrical characteristics and sensitivity features into a preset system coordination and optimization unit, and calculate the optimal operating point and power allocation command of DC power supply with grid quality, load balance and system efficiency as collaborative objectives. The adaptive execution module is used to generate network coordination control commands based on the optimized operating point and power allocation instructions, and send them to the distribution management unit or load switch connected to the DC power supply to adjust the output power of the DC power supply in the distribution network and its coordination relationship with other power supplies.

[0015] This embodiment provides a DC power supply coordination and control system based on industrial big data, which aims to solve the problem that the existing DC power supply control only focuses on electrical indicators and ignores the disconnect between the physical state of the load side and the final industrial production quality. The specific execution process of the system is as follows: The system performs a full-domain data acquisition task through a multi-dimensional heterogeneous data sensing module. This module, as the system's sensor, not only collects electrical quantities on the distribution network side through high-frequency sensors, but also deeply integrates industrial fieldbus to collect non-electrical quantities on the load side, breaking down the data barriers between the electrical domain and the physical domain. Based on this, the feature mapping module, as the cognitive center of the system, receives operating status data and power demand data, and uses deep learning algorithms to mine the physical state of the load that is difficult to observe directly, namely the hidden physical layer, which specifically refers to the real-time thermodynamic distribution, mechanical stress accumulation state and electrochemical reaction activity of the load equipment. It constructs a load characteristic and power quality correlation model, and does not directly output voltage commands, but outputs the dynamic electrical characteristics and sensitivity characteristics of the load. The multi-objective collaborative decision-making module acts as the decision-making brain, inputting the aforementioned dynamic electrical characteristics and sensitivity features into the preset system coordination and optimization unit. Its decision-making logic is no longer a simple constant voltage or constant current, but rather to find an optimal operating point that optimizes the production process indicators on the load side while satisfying the power grid quality constraints. The adaptive execution module acts as the system's execution hand, generating network coordination control commands based on the optimized operating point and power allocation instructions, and sending them to the power distribution management unit or load switch connected to the DC power supply. This embodiment achieves a leap from equipment-level electrical index control to production line-level process power supply collaborative control by constructing a perception and decision-making architecture that spans the electrical and physical domains. Production line-level control refers to a control mode in which the control objective function includes industrial production process indicators other than electrical indicators. The system can effectively identify load physical state changes that cannot be perceived by simple electrical control, such as impedance drift or thermal inertia hysteresis, and maximize the production yield of industrial products and system energy efficiency while ensuring grid quality by adjusting the collaborative relationship of DC power supply.

[0016] Example 2: The process by which the multidimensional heterogeneous data sensing module collects operational status data from the distribution network side and electricity demand data from the load side includes: acquiring voltage, frequency, and harmonic distortion rate data at the distribution network's point of common coupling, as well as output current, power factor, and available capacity data of the DC power supply itself, marking them as operational status data, and setting a millisecond-level acquisition frequency; acquiring planned production curves, real-time active / reactive power demand, and voltage tolerance range data of key loads from the load side, marking them as electricity demand data, and setting a second-level acquisition frequency; and aligning the operational status data and electricity demand data collected at the same time in time to generate a multidimensional heterogeneous dataset.

[0017] This embodiment further specifies the data acquisition and processing of the multidimensional heterogeneous data sensing module, focusing on solving the alignment problem of multi-physics data in the spatiotemporal dimension; the specific steps are as follows: the multidimensional heterogeneous data sensing module acquires data of the common connection point of the distribution network and the DC power supply body through high-precision power monitoring instruments. The acquisition objects cover voltage, frequency, harmonic distortion rate, as well as the output current, power factor and available capacity of the DC power supply, and the sampling frequency is set to high-frequency mode to capture electrical transients; The module acquires load-side data through the industrial control system interface, including planned production curves, real-time active and reactive power demand, and voltage tolerance range of critical loads. The sampling frequency is set to a low-frequency mode to match the time constants of thermodynamic or chemical reactions. To eliminate data phase differences caused by different sampling frequencies, this embodiment uses timestamp anchoring and interpolation algorithms to time-align the operating status data and power demand data collected at the same time, constructing a multi-dimensional heterogeneous dataset. To meet the causal constraints of the real-time control system, this embodiment uses a first-order linear extrapolation algorithm instead of non-causal interpolation. The specific calculation formula is as follows: in, The source is extrapolated calculation, and the physical meaning is at the electrical sampling time. Estimated process data; The source is historical low-frequency sampling, and the physical meaning is time. The most recent raw low-frequency process data point; The source is historical low-frequency sampling, and the physical meaning is time. The preceding adjacent original low-frequency process data points; The source is the system clock, and its physical meaning is the sampling time of the most recent raw low-frequency data. The source is the system clock, and its physical meaning is the sampling time of the previous raw low-frequency data. This embodiment effectively integrates fast variables of electrical quantities and slow variables of process quantities on the same time axis through a hierarchical acquisition and time-series alignment mechanism. This method provides an accurate data foundation for subsequent exploration of the lagging impact of voltage fluctuations on product quality, avoids misjudgment of causal relationships caused by data asynchrony, and thus improves the fidelity of multi-source data fusion.

[0018] Example 3: The process of constructing the load characteristic and power quality correlation model by the feature mapping module includes: acquiring historical multidimensional heterogeneous datasets as training samples, extracting the trend of electricity demand data changes in the training samples as input features, and extracting the correlation between the corresponding load-end voltage deviation and key process parameter fluctuations in the training samples as output labels; constructing an initial correlation model using a long short-term memory network algorithm, iteratively training the initial correlation model using training samples, and obtaining the completed load characteristic and power quality correlation model; inputting real-time collected electricity demand data into the load characteristic and power quality correlation model, and outputting the load dynamic electrical characteristics and sensitivity features under the current operating conditions.

[0019] This embodiment is a further specification of the feature mapping module's construction of a load characteristic and power quality correlation model, which is the core step of this invention to identify the hidden physical layer; the specific execution is as follows: obtain historical multidimensional heterogeneous datasets as training samples. The data comes from the operation logs of the power distribution network SCADA system and industrial energy management system over the past 24 months, containing 50,000 sets of cleaned and time-aligned valid samples. Extract the trend of electricity demand data changes as input features, and extract the correlation between the corresponding load-end voltage deviation and key process parameter fluctuations as output labels; To address the issue that association relationships cannot be trained as neural network labels, this embodiment defines a specific label quantization generation function: Let at time... Preset window Key process parameters, such as electrolytic cell temperature or coating thickness, have actual values. The standard threshold allowed by the process is Then output the label. Construct a normalized risk probability value: in, The source is a preset parameter, and its physical meaning is the length of the look-ahead time window used to evaluate process stability, in seconds; The source is real-time sensor readings, and the physical meaning is in the future. The actual values ​​of key process parameters are determined by the specific process, such as degrees Celsius or micrometers. The source is the process specification, and the physical meaning is the target reference value of the process parameter, i.e., the set point, with the same unit as above; The source is the quality control standard, and its physical meaning is the maximum threshold value that can be deviated from the reference value. The unit is the same as above. The source is the algorithm hyperparameter, and its physical meaning is the steepness coefficient. For example, a value of 10 is used to control the sensitivity of the probability mapping. The unit is dimensionless. This formula maps the parameter fluctuations of the physical layer to sensitivity probabilities in the [0,1] interval, enabling the model to learn a nonlinear mapping from electrical characteristics to process risks. Considering the memory nature of industrial loads, this embodiment uses the Long Short-Term Memory (LSTM) network algorithm to construct the initial correlation model, and uses training samples to iteratively train the initial correlation model. To accurately capture the long-range dependencies of time series, the specific update formula for the LSTM unit is defined as follows: in, : These represent the activation vectors of the forget gate, input gate, and output gate, respectively, with values ​​ranging from [0,1], and are used to control the flow of information; : Represents the current cell state and candidate cell state respectively. Physically, it represents the long-term memory characteristics of the load, carrying the cumulative influence of historical conditions. The source is model calculation, the physical meaning is the load hidden state feature vector at the current moment, and the unit is dimensionless; Specifically, it is constructed as a standardized vector consisting of the current active power demand, reactive power demand, and node voltage deviation: in, The source is the sensor input value after Z-Score normalization, and the physical meaning is the dimensionless electrical and demand characteristics input at the current moment; The unit depends on the input quantity; The source is the model parameters, and the physical meaning is the weight matrix of the corresponding gate; The source is the model parameters, and the physical meaning is the bias term of the corresponding gate; The source is a preset function, namely the Sigmoid activation function and the hyperbolic tangent activation function; : Represents the Hadamard product of a matrix; During the inference phase, the trained model takes real-time collected electricity demand data as input and outputs the dynamic electrical characteristics of the load under the current operating conditions, i.e., the sensitivity feature value, through a fully connected output layer function. To clarify the specific implementation method, the function... Defined as a combination of a linear transformation layer and a Sigmoid activation function, the formula is as follows: in, The source is the model parameters, and the physical meaning is the weight matrix of the output layer, with dimensions adapted to the hidden state and output node; The source is the model parameters, and its physical meaning is the bias vector of the output layer; The source is the model output, and the physical meaning is the sensitivity feature value, which is used to quantify the probability that a voltage fluctuation will cause the process parameters to exceed the standard at the current moment. The unit is the normalized value. Furthermore, in order to provide the necessary load dynamic electrical characteristics for power flow optimization operations, i.e., the ZIP model coefficient input, the LSTM model in this embodiment, in addition to outputting sensitivity features... In addition, a second output branch, namely a regression layer, is set up in parallel to predict the instantaneous ZIP coefficient of the load; the specific calculation formula of this regression layer is as follows: in, These correspond to the constant impedance, constant current, and constant power coefficients in the load model, respectively. The function guarantees that the sum of the three is 1, which complies with physical constraints; The weights and biases of this regression branch are defined. To address the training label problem for this physical parameter branch, this embodiment introduces a recursive least squares (FF-RLS) recognizer with a forgetting factor during the training preprocessing stage: targeting historical voltage... With power Sequence, constructing regression vectors: Iteratively update the parameter estimation vector using the following steps. Excitation Dead Zone Detection: To avoid singular covariance matrices due to a lack of discriminative power in the input vector under steady-state voltage conditions, the system presets a voltage change dead zone threshold. ; Determine if the parameter update conditions are met: If the above conditions are not met, then directly set And skip the subsequent steps, preserving the covariance matrix. If the condition is met, continue with the following steps: Update parameter estimates: Update the covariance matrix: Label normalization: Since the output layer of the LSTM network uses the Softmax function to force the sum of coefficients to be 1, in order to ensure the logical consistency between the training labels and the network output, the RLS estimates need to be normalized using the L1 norm. in, Set to 0.98 to accommodate the time-varying nature of the load characteristics; eventually converges and is normalized. The value is used as the GroundTruth label for supervised learning of the LSTM network, thereby giving the model the ability to infer physical characteristics from demand trends; Building upon this, in order to resolve the gradient conflict problem between different output branches in multi-task learning and to clarify the training objective, this embodiment discloses a specific form of the composite loss function: in, : This is the binary cross-entropy loss function, used to optimize sensitivity features. The calculation formula is: : This refers to the Kullback-Leibler divergence, or KL divergence loss function for short, which measures the distribution of predicted ZIP coefficients. The true distribution generated by RLS The difference between them is calculated using the following formula: KL divergence was chosen because the ZIP coefficients, after being processed by Softmax, exhibit probability distribution characteristics. : represents the task weight coefficient, which is the gradient magnitude for balancing classification and regression tasks. ; : This is the L2 regularization term, used to prevent overfitting; this loss function ensures that the model can simultaneously take into account the accuracy of process sensitivity identification and the precision of physical parameter regression during backpropagation; This embodiment utilizes the powerful timing modeling capabilities of the LSTM algorithm to construct a digital twin of the load. This model can dynamically identify the dynamic characteristics of the load under different operating conditions. For example, it can identify that the electrolytic cell is extremely sensitive to harmonics at a specific temperature, while it is robust at other temperatures, thus providing a quantitative basis for subsequent precise control.

[0020] Example 4: The process by which the multi-objective collaborative decision-making module calculates the optimal operating point and power allocation command for the DC power supply includes: constructing a system coordination optimization unit, which includes distribution network operation constraints, multi-DC power supply coordination rules, and load priority strategies; extracting the real-time operating conditions of the power grid and the DC power supply status from the operating status data, and combining the dynamic electrical characteristics and sensitivity features of the load, and inputting them into the system coordination optimization unit; the system coordination optimization unit performs optimization calculations with the objectives of minimizing the voltage and frequency deviation at the point of common coupling, minimizing the variance of the load rate of each power supply, and prioritizing the power supply to critical loads, and outputs the optimal operating point for the DC power supply, including the target output power, power factor, and power allocation command.

[0021] This embodiment further specifies the solution for optimizing the operating point using a multi-objective collaborative decision-making module. The specific steps are as follows: A system coordination optimization unit is constructed. This unit acts as a mathematical programming solver, integrating node voltage constraints, line thermal stability constraints, and current sharing or capacity-based allocation rules among multiple DC power sources. Real-time grid operating conditions and DC power source states are extracted from the operating status data, and combined with the load's dynamic electrical characteristics and sensitivity features, and input into this unit. Specifically, the combination operation is not a simple superposition but is achieved through a vector concatenation operator: Constructing the system state input vector: in, This is the power grid operating condition vector. It is the power supply state vector. For sensitivity feature scalar, The load power vector is used as the basis, and each component is normalized to the [0,1] interval by Min-Max before splicing, thus clarifying the input data structure of the optimizer. Based on this, a multi-objective weighted cost function is constructed to balance grid quality, load balance, and system efficiency. In order to solve the problem that the optimization weights fail due to large differences in the values ​​of different physical quantities, such as voltage, frequency, and power, this embodiment normalizes the various indicators. The cost function formula is as follows: in, The source is a calculated value, the physical meaning is the overall operating cost of the system, and the unit is dimensionless; The source is the system topology configuration list, and the physical meaning is the start time of this optimization calculation cycle. The total number of all online and normally communicating voltage monitoring nodes within the distribution network segment; The source is the operational status feedback, and its physical meaning is the total number of controllable power supply units currently in grid-connected operation. The system locks this set before each round of optimization. If node communication is interrupted during the calculation process, the initial snapshot is still used as the standard, and missing data is filled in by state estimation to determine the unique calculation range. The source is real-time monitoring, and the physical meaning is a node. The real-time voltage, in volts; The source is a preset value, the physical meaning is reference voltage, and the unit is volts; The source is real-time monitoring, and the physical meaning is the real-time frequency of the common connection point, measured in Hertz. The source is a preset value, and its physical meaning is the system's rated frequency, measured in Hertz. The source is equipment registration information, and the physical meaning is the total number of DC power supply modules participating in the collaborative control. The source is a calculated value, and the physical meaning is the first. The load factor of a DC power supply is defined by the following formula: in, The corresponding decision variable in the optimization algorithm, i.e., the first The output power setting value of each power supply, in percentage; The source is a calculated value, and its physical meaning is the average load rate of all power sources. The calculation formula is: The unit is a percentage; The source is operational status feedback, and its physical meaning is the total number of controllable power supply units currently in grid-connected operation. The source is equipment parameters, and the physical meaning is the first... The rated power of the power supply, used for normalization of the power term, is expressed in watts. The source is equipment parameters, and the physical meaning is the first... Each power supply has an output power The conversion efficiency function is set to a decimal form (0-1) to ensure dimensionality consistency in the denominator calculation. This embodiment specifically uses a quadratic polynomial model to describe this efficiency characteristic. Note that the variables entered in the formula need to be converted to per-unit values. To adapt the fitting coefficients ; The source is a preset constant, and its value is... Used to prevent power outages, i.e. When the denominator is zero, the calculation overflows; this term Physically equivalent to minimizing the total input power of the system, i.e. maximizing overall energy efficiency; The source is the preset weight, and its physical meaning is the weight coefficient of each optimization objective, with the unit being dimensionless; in this embodiment, the recommended baseline weight value is: The specific method for obtaining the weights is as follows: a judgment matrix is ​​constructed using the Analytic Hierarchy Process (AHP), and a set of benchmark weights is calculated. Furthermore, in order to enable the optimization objective to respond to the sensitivity characteristics output by the preceding module, this embodiment introduces a dynamic weight adjustment mechanism: in, The baseline weight for the voltage deviation target. The real-time sensitivity feature output by the feature mapping module, with values ​​ranging from 0 to 1. This is a preset sensitivity gain coefficient, for example, set to 5.0; the physical meaning of this formula is: when the load is in a highly sensitive state, i.e. At that time, the system automatically and significantly increases the weight of voltage stability, thereby achieving the technical effect of prioritizing the power supply of critical loads at the optimization level; Specifically, in the above cost function During the iterative calculation process, node voltage and frequency It is not a fixed value, but varies with the output power of the DC power supply. The dependent variable changes; to achieve closed-loop computation in the particle swarm optimization algorithm, this embodiment embeds a linearized Jacobian sensitivity model for quickly estimating the power grid state corresponding to candidate solutions: in, The voltage-power sensitivity coefficient is obtained and updated according to the following logic: The system optimization cycle is set. The trigger time is 15 minutes, or immediately when the total load fluctuation rate of the distribution network exceeds 5%; it is triggered at the beginning of each optimization cycle, i.e., at the particle swarm iteration. At this time, the system triggers a baseline power flow calculation, generating a Jacobian matrix based on the currently collected power grid topology and operating point. And by solving the system of linear equations The voltage-power sensitivity coefficient can be directly calculated. This avoids the waste of computational resources caused by directly inverting large sparse matrices; During the particle swarm iteration process, i.e. to ,Should The parameters remain constant until the start of the next new optimization cycle, at which point they are updated again. This mechanism ensures that the parameters are derived from a snapshot of the state at the start of the optimization cycle, thus guaranteeing model consistency during each optimization process. This is the frequency droop coefficient of the system; In addition, the other parameters involved in the formula are defined as follows: The source is the algorithm iteration variable, and its physical meaning is the value at the first iteration. In the nth iteration calculation, the th A DC power supply, corresponding to the particle dimension The candidate output power setting value, in watts; The source is real-time sensor data acquisition; its physical meaning is that at the start of the optimization calculation, the [missing information]... The current actual output power measurement of a DC power supply is used as the reference operating point for linearization, in watts; The source is a multi-dimensional heterogeneous data sensing module, and its physical meaning is the measured value of the total active power demand of the distribution network segment at the current moment, in watts; This model ensures that the optimization process can perceive the actual impact of power regulation on grid quality; and calculates the new position in each iteration. Then, the corresponding cost function value needs to be calculated. ,like Then update ;like Then update ; To ensure the feasibility and convergence speed of the algorithm, this embodiment uses an improved particle swarm optimization (PSO) algorithm to solve the above cost function. The minimum value; to fully disclose the calculation process, the specific iterative formula for particle update is defined as follows: To prevent particles from flying out of the physically feasible region, a boundary clamping operation must be performed after updating the position: in, : No. In the next iteration, particles In dimensions The corresponding power and speed of each power source; : No. In the next iteration, particles In dimensions The location of the power distribution value to be solved; Minimum and maximum output power limits for DC power supplies; The inertia weight is set to decrease linearly from 0.9 to 0.4 to balance global search and local exploitation capabilities. The learning factor is set to 2.0 for all. : A random number in the range [0,1]; The best position in the history of a particle, that is, the coordinates at which the particle has the highest fitness in each iteration; The global optimal position of the population is the coordinate with the highest fitness in the entire particle swarm throughout the iterations. The algorithm iterates until the preset maximum number of iterations is reached or the fitness value no longer decreases significantly. The output of the global optimal position at this time is the optimized operating point and power allocation command of each DC power supply. This embodiment no longer pursues the local optimum of a single power supply, but achieves the global optimum at the system level through multi-objective weighted optimization; in particular, it introduces a load rate variance minimization term, which effectively prevents some power supplies from being overloaded while others are lightly loaded, thereby extending the average lifespan of the power supply system and improving the overall energy efficiency.

[0022] Example 5: The multi-objective collaborative decision-making module is also used to perform multi-source collaborative and load dispatch operations, which include: preset power grid quality safety range; obtain real-time voltage or frequency data of the current point of common coupling and predict the voltage or frequency value at the next moment based on the rate of change; determine whether the real-time data and the predicted voltage or frequency value at the next moment are within the safety range; if the real-time data and the predicted voltage or frequency value at the next moment do not deviate from the safety range, then keep the current output command of each DC power source unchanged. If the real-time data is within the safe range but the predicted voltage or frequency value for the next moment deviates from the safe range, it is determined that a deviation is imminent. The collaborative scheduling program is then initiated to recalculate the output combination of each controllable unit in the system, including DC power supplies and other adjustable distributed power supplies / energy storage, and to generate new optimized operating points and power allocation instructions. The instructions are simultaneously sent to multiple controlled units.

[0023] This embodiment further specifies the multi-source coordination and load dispatching operation, aiming to solve the dynamic adjustment problem when the system is about to deviate from the safe range. The specific process is as follows: a power grid quality safe range is preset, and real-time voltage data of the current point of common coupling is obtained; the system predicts the voltage value at the next moment based on the voltage change rate, and the prediction formula is as follows: in, The source is a predictive calculation, and its physical meaning is the predicted voltage value at the next moment, with the unit being volts; The source is real-time sampling, and its physical meaning is the voltage sampling value at the current moment, with the unit being volts; The source is differential calculation, and its physical meaning is the rate of change of voltage, with the unit being volts per second; The source is a preset value, and its physical meaning is the prediction step size, with the unit being seconds. The system determines whether the real-time data and the predicted voltage value for the next moment are within the safe range; in response to the predicted value deviating from the safe range, the system determines that it is about to deviate and immediately starts the collaborative scheduling program; it recalculates the output combination of each controllable unit in the system, generates a new optimized operating point and power allocation command, and sends it to multiple controlled units at the same time. This embodiment introduces a prediction mechanism based on the rate of change, transforming the control mode from traditional lag correction to proactive prevention. This strategy is crucial for voltage-sensitive precision industrial loads, effectively preventing production interruptions or scrap caused by adjustment lag, and significantly improving the system's proactive safety.

[0024] Example 6: The system includes multiple DC power sources connected to the same distribution network segment. The multi-objective collaborative decision-making module is also used to perform network power flow optimization operations, which include: acquiring the topology of the distribution network segment, the impedance parameters of each line, and the demand forecast of all load nodes; predicting the network power flow distribution and node voltage levels under different power output schemes based on the dynamic electrical characteristics and sensitivity features of the load; and solving for the optimal power flow with the goal of minimizing network losses and balancing line load rates, thereby obtaining the optimal operating point and power allocation instructions for each DC power source.

[0025] This embodiment further specifies the network power flow optimization operation performed by the multi-objective collaborative decision-making module, and is particularly suitable for scenarios where multiple DC power sources are connected to the same distribution network segment. The specific execution logic is as follows: The topology of the distribution network segment, the impedance parameters of each line, and the demand forecasts for all load nodes are obtained; the system stores the node admittance matrix of the distribution network segment; based on the dynamic electrical characteristics of the loads combined with Kirchhoff's laws, the network power flow distribution and node voltage levels under different power output schemes are predicted; to specifically implement the constraint based on dynamic electrical characteristics, the load node power in this embodiment... It is no longer considered a constant value, but is described using a voltage-dependent ZIP model: in, The source is a load database, and the physical meaning is a node. The nominal active power at rated voltage, expressed in kilowatts (kW). The source is the power grid standard, and its physical meaning is the rated voltage of the distribution network, with the unit being volts (V). The source is the output of the feature mapping module, and its physical meaning is the constant impedance component coefficient, which characterizes the sensitivity of the load to the square of the voltage. The unit is dimensionless. The source is the output of the feature mapping module, and its physical meaning is the constant current component coefficient, which characterizes the linear sensitivity of the load to voltage. The unit is dimensionless. The source is the output of the feature mapping module, and its physical meaning is the constant power component coefficient, which characterizes the load's characteristics independent of voltage. The unit is dimensionless. These coefficients are assigned in real-time by the aforementioned feature mapping module based on the current load characteristics identification results; to clarify the specific implementation of real-time matching and assignment, this embodiment adopts shared memory mapping technology: the feature mapping module calculates the coefficients... Write the values ​​to a high-speed key-value pair memory database or shared memory area indexed by the load ID. The optimization module can directly obtain the latest coefficients through key-value query operations. If the query times out, the value of the previous period will be used by default. To strictly satisfy physical constraints during the interior-point method solution process, this embodiment constructs nodal power imbalance equations incorporating ZIP parameters as equality constraints. : in, Inject power into the power source. Let be the real and imaginary parts of the nodal admittance matrix elements. The load power factor angle; this set of equations clarifies the load dynamic characteristics. How to couple network power flow constraints; construct an optimization function with the goal of minimizing network loss and balancing line load: in, This represents the set of all lines within a distribution network segment; The baseline capacity of the system, for example, taking the rated capacity of the transformer in the distribution network section as 10MVA, is used to account for active power losses. Normalization to per-unit values ​​resolves the dimensional inconsistency between this value and the second term, the dimensionless load factor variance, when directly weighted and summed; the first term... The total network loss is calculated as follows: The second item is the line load balancing index, among which... Indicates the connection node Real-time line load rate, This represents the average load rate of the entire network. These are the corresponding weighting coefficients; this objective function ensures that the current will not concentrate on a few low-impedance lines, leading to thermal overload. To ensure the physical feasibility of the power flow solution, the optimization model must also satisfy the following inequality constraints: in, Branch conductance, derived from the distribution network topology parameter library, is measured in Siemens s. Node voltage amplitude; Node voltage phase angle difference; The safe upper and lower limits of node voltage are usually set to the rated voltage. ; The corresponding formula ,:Line apparent power flow, calculated from power flow, in kilovolt-amperes (kVA); The corresponding formula ,:Line thermal stability limit capacity, derived from line design specifications, unit is kilovolt-amperes (kVA); For the aforementioned nonlinear programming problem, this embodiment employs the primal-dual interior-point method for solution. By introducing a logarithmic barrier function, the inequality constraints are transformed into equality constraints, and the corrected equations are calculated iteratively using Newton's method. The calculation of the Newton direction depends on the Jacobian matrix of the aforementioned power imbalance equations. To ensure convergence under dynamic load characteristics, this embodiment clarifies the correction formula for the load derivative term included in the diagonal elements of the Jacobian matrix: The aforementioned derivative terms rigorously correct for the error in traditional power flow calculations that assume the partial derivative of load power with respect to voltage is zero, ensuring the algorithm's convergence speed under ZIP loads; until the complementary gap is less than the preset tolerance, such as This allows us to derive the optimal operating point and power allocation instructions for each DC power supply. This embodiment solves the circulating current problem and line loss problem when multiple power sources are connected in parallel by introducing network topology constraints and optimal power flow calculation. The scheme ensures that power is not only properly allocated at the power source end, but also that the loss is minimized during transmission, thus achieving true source-network-load coordinated optimization.

[0026] Example 7: The process by which the adaptive execution module generates network coordination control commands based on the optimized operating point and power allocation commands includes: acquiring the current upper-level scheduling command or local setpoint received by the DC power supply; calculating the absolute value of the deviation between the optimized operating point and power allocation command and the current command; determining whether the absolute value of the deviation is greater than a preset dead zone threshold; if it is greater than the dead zone threshold, generating a control command sequence based on the deviation, and issuing the control commands through the power monitoring network or energy management system communication protocol. The control objective is to adjust the output power setpoint of the DC power supply or operate the associated grid-connected switch and power regulation unit.

[0027] This embodiment further specifies the network coordination control instructions generated by the adaptive execution module, particularly by introducing dead-zone control to avoid oscillations. The specific process is as follows: Obtain the current upper-level scheduling instruction or local setpoint received by the DC power supply, and calculate the absolute value of the deviation between the optimized operating point and power allocation instruction and the current instruction; set an adjustment dead-zone threshold, which is not a fixed value but dynamically adjusted according to load sensitivity characteristics; the dead-zone threshold calculation formula is as follows: in, The source is dynamic calculation, and the physical meaning is the adjustment dead zone threshold, i.e., the minimum response amplitude, and the unit is ampere or watt. The source is a preset value, and its physical meaning is the reference dead zone coefficient. In this embodiment, it is specifically set to 2% of the rated output power of the DC power supply, that is... To balance the adjustment of sensitivity and the device's anti-shake function; The source is the preceding calculation, the physical meaning is the load sensitivity characteristic, and the unit is a normalized value; it determines whether the absolute value of the deviation is greater than the above-mentioned dead zone threshold; in response to the deviation being greater than the threshold, a control command sequence is generated based on the deviation; in order to eliminate the black-box description of command generation based on deviation and ensure the smoothness of command generation, this embodiment uses an incremental PID algorithm to generate specific power regulation commands: in, The power deviation value calculated above, , These are the power deviation values ​​for the previous control cycle and the two previous control cycles, respectively. Set an increment for the power generated in this adjustment cycle. The preset discrete control gain; specifically: , in, To control the cycle, These are the simulation domain coefficients; in this embodiment, The values ​​are 0.8, 0.2, and 0.05 respectively, which already include time constant adjustments; the system then calculates the new absolute setpoints. Then, it is encapsulated into a Modbus-TCP or IEC61850 message; the instruction is issued through the power monitoring network or energy management system communication protocol to directly modify the duty cycle setting of the DC power supply or switch the grid connection switch status. This embodiment introduces a dynamic dead-zone control mechanism, which cleverly resolves the contradiction between control accuracy and equipment lifespan. When the load is not sensitive, coarse control is used to reduce switching losses and equipment wear, while fine control is used when the load is sensitive to ensure quality. This achieves adaptive and flexible switching of the control strategy and enhances the robustness of the system.

[0028] Example 8: The adaptive execution module also includes a predictive power quality management unit, which performs the following operations: based on historical and real-time operating status data, predicts the voltage sag, harmonic distortion rate increment, or power shortage that may occur in the distribution network within a preset time period; if the predicted event will affect critical loads and the voltage sag, harmonic distortion rate increment, or power shortage exceeds a preset severity threshold, it generates a preventive control instruction set in advance; in response to this instruction set, it coordinates the response strategies of DC power supplies, energy storage systems, and reactive power compensation devices to adjust the network operating status before the event occurs, thereby achieving proactive defense.

[0029] This embodiment further specifies the predictive power quality management unit of the adaptive execution module, performing proactive defense operations. The specific steps are as follows: Based on historical and real-time operating status data, an anomaly detection model is used to predict power quality indicators within a preset time period. The anomaly detection model is specifically constructed as a sequence prediction network based on a gated recurrent unit (GRU). The network input layer receives voltage, current, and frequency time series from the past N cycles. Considering the significant differences in the dimensions of voltage (V), current (A), and harmonic distortion rate (%), direct input would cause the network gradient to vanish or explode. Therefore, Z-score normalization processing must be performed before inputting into the GRU unit. in, For the first Dimensional features in The original value at time, and These are the mean and standard deviation of this dimension within the sliding window, respectively; the standardized data are input into the GRU cell, and the specific calculation formula is as follows: in, The updated gate determines how much historical information is retained; Reset the gate to determine how much historical information to ignore; : Candidate hidden state; Output the hidden state at the current moment; The corresponding weight matrix; after extracting time-series features through a multi-layer GRU, the voltage amplitude sequence for the next M periods is output by a fully connected output layer function. Harmonic content prediction and total load power demand sequence To enable those skilled in the art to reproduce this multi-step sequence generation process, the calculation formula for the output decoding layer is defined as follows: In addition, to predict power demand, the model is configured with a parallel power decoding branch: in, For voltage decoding weight matrix, Hidden state vector Dimensions The ReLU activation function is used to ensure the non-negativity of the power prediction value, with weights and biases for the power branch. This layer directly maps the hidden states of the GRU at time steps to the predicted values ​​for the next M time steps. To transform the predicted sequence into a comparable scalar indicator, the system extracts the worst-case scenario in the sequence as the predicted value and calculates the power shortage. in, The severity index is the sum of the maximum available capacity of all power supplies in the current system; the formula is as follows: in, The source is a calculated value, the physical meaning is a severity index, and the unit is dimensionless; The source is the sequence extracted value, and its physical meaning is the predicted voltage sag amplitude, with the unit being volts; The source is equipment parameters, the physical meaning is rated voltage, and the unit is volts; The source is the predicted value, and its physical meaning is the predicted increase in harmonic distortion rate, expressed as a percentage. The source is a predictive calculation; its physical meaning is the predicted power shortage; the unit is watts. The source is system parameters, and the physical meaning is the total design capacity of the system, used for normalization; The source is a preset value, and its physical meaning is the normalized weight coefficient, with the unit being dimensionless; at the same time, in order to obtain the predicted value of harmonic distortion rate, the model is configured with a third output branch: in, The harmonic decoding weight matrix has an output range constrained within... Interval; based on this sequence, calculate the predicted harmonic distortion rate increment: in, The measured harmonic values ​​are given at the current moment. To enable those skilled in the art to obtain the GRU model with multidimensional prediction capabilities, this embodiment further discloses the training and construction process of the model: constructing a training sample set containing historical voltage sequences, current sequences, and corresponding future measured values ​​of voltage, power, and harmonics; defining a multi-task composite loss function. To guide model parameters The iterative update is performed using the following formula: in, The future of the training samples The measured true values ​​of voltage and power at each moment; This is a binary label for harmonic distortion exceeding the standard. If the measured harmonic distortion rate exceeds the standard, it is 1; otherwise, it is 0. This represents the mean squared error loss (MSE), used in regression tasks. This represents the binary cross-entropy loss, used for classification tasks; To balance the weights of the gradients for each task, these are set to 0.5 and 1.0 respectively in this embodiment; the Adam optimizer is used, with an initial learning rate of... The model is trained by backpropagation until the loss function converges, thereby determining the specific values ​​of each weight matrix. The system determines whether the severity index exceeds a preset threshold. This preset threshold is not a fixed value but is set based on the ITIC (Information Technology Industry Council) voltage tolerance curve or the SEMIF47 semiconductor equipment voltage sag immunity standard. If the predicted point falls into the prohibited area of ​​the standard curve, it is determined that the threshold has been exceeded. In response to the index exceeding the limit, it determines that the predicted event will affect critical loads and generates a preventative control command set in advance. In response to this command set, it coordinates the response strategies of the DC power supply, energy storage system, and reactive power compensation device. The specific calculation logic of this response strategy is as follows: Calculate the reactive power compensation increment command. Based on predicted voltage sag Instructions are generated using piecewise linear functions: in, To preset the compensation gain, for example, 2.5 kvar / V, The dead zone voltage is, for example, 3V; this command is issued to the reactive power compensation device to establish reactive power support before a voltage dip occurs; the active power support command is calculated. : in, This represents the current remaining capacity of the DC power supply. The advance support coefficient is set to 1.1; the instruction is sent to the DC power supply execution unit; multi-device collaborative arbitration logic: to address possible instruction conflicts during the coordination process, such as the DC power supply being unable to fully respond when fully loaded. If the energy storage system is currently charging, this embodiment defines a priority-based arbitration rule: power quality management is prioritized over economic operation; when a preventative control command set is generated, the system forcibly interrupts the energy storage system's charging task and switches it to discharge mode to fill the power gap in the DC power supply, ensuring... To obtain full compensation; to adjust the network's operating status before an incident occurs, thus achieving proactive defense.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A DC power supply coordination control system based on industrial big data, characterized in that, include: In the cloud, the cloud communication connection includes a multi-dimensional heterogeneous data perception module, a feature mapping module, a multi-objective collaborative decision-making module, and an adaptive execution module; The multidimensional heterogeneous data sensing module is used to collect operating status data on the distribution network side and electricity demand data on the load side. The feature mapping module is used to construct a load characteristic and power quality correlation model based on the operating status data and power demand data, and output the dynamic electrical characteristics and sensitivity characteristics of the load. The multi-objective collaborative decision-making module is used to input the dynamic electrical characteristics and sensitivity features into a preset system coordination and optimization unit, and calculate the optimal operating point and power allocation command of the DC power supply with grid quality, load balance and system efficiency as collaborative objectives. The adaptive execution module is used to generate network coordination control instructions based on the optimized operating point and power allocation instructions, and send them to the power distribution management unit or load switch connected to the DC power supply, so as to adjust the output power of the DC power supply in the power distribution network and its coordination relationship with other power supplies.

2. The DC power supply coordination control system based on industrial big data according to claim 1, characterized in that, The process by which the multidimensional heterogeneous data sensing module collects operating status data from the distribution network side and electricity demand data from the load side includes: Acquire the voltage, frequency, and harmonic distortion rate data of the distribution network common connection point, as well as the output current, power factor, and available capacity data of the DC power supply body, mark them as operating status data, and set the millisecond-level acquisition frequency; Acquire the planned production curve, real-time active / reactive power demand, and voltage tolerance range data of key loads on the load side, mark them as power demand data, and set the second-level acquisition frequency; The operational status data and electricity demand data collected at the same time are time-series aligned to generate a multidimensional heterogeneous dataset.

3. The DC power supply coordination control system based on industrial big data according to claim 2, characterized in that, The process by which the feature mapping module constructs a model relating load characteristics and power quality includes: Historical multidimensional heterogeneous datasets are obtained as training samples. The trend of electricity demand data change in the training samples is extracted as input features. The correlation between the load-end voltage deviation and the fluctuation of key process parameters in the corresponding training samples is extracted as output labels. An initial correlation model is constructed using a long short-term memory network algorithm. The initial correlation model is then iteratively trained using training samples to obtain a completed correlation model between load characteristics and power quality. The real-time collected electricity demand data is input into the load characteristics and power quality correlation model, and the dynamic electrical characteristics and sensitivity features of the load under the current operating conditions are output.

4. The DC power supply coordination control system based on industrial big data according to claim 3, characterized in that, The process by which the multi-objective collaborative decision-making module calculates the optimal operating point and power allocation command for the DC power supply includes: A system coordination and optimization unit is constructed, which includes distribution network operation constraints, multi-DC power source coordination rules, and load priority strategies. Extract the real-time operating conditions of the power grid and the DC power supply status from the operating status data, and input them into the system coordination and optimization unit in combination with the dynamic electrical characteristics and sensitivity features of the load. The system coordination and optimization unit performs optimization calculations with the goal of minimizing the voltage and frequency deviation at the point of common coupling, minimizing the variance of the load rate of each power supply, and prioritizing the power supply to critical loads. It outputs the optimized operating point of the DC power supply, including the target output power, power factor, and power allocation instructions.

5. A DC power supply coordination control system based on industrial big data according to claim 4, characterized in that, The multi-objective collaborative decision-making module is also used to perform multi-source collaboration and load scheduling operations, which include: Preset the power grid quality safety range; Obtain real-time voltage or frequency data of the current point of common coupling, and predict the voltage or frequency value at the next moment based on the rate of change; Determine whether the real-time data and the predicted voltage or frequency value for the next moment are within the safe range; If the real-time data and the predicted voltage or frequency value for the next moment do not deviate from the safe range, then the current output command of each DC power supply remains unchanged. If the real-time data is within the safe range but the predicted voltage or frequency value for the next moment deviates from the safe range, it is determined that a deviation is imminent. The collaborative scheduling program is then initiated to recalculate the output combination of each controllable unit in the system, including the DC power supply and other adjustable distributed power sources / energy storage, and to generate new optimized operating points and power allocation instructions. These instructions are simultaneously sent to multiple controlled units.

6. The DC power supply coordination control system based on industrial big data according to claim 1, characterized in that, The system includes multiple DC power sources connected to the same distribution network segment. The multi-objective collaborative decision-making module is also used to perform network power flow optimization operations, which include: Obtain the topology of the power distribution network segment, the impedance parameters of each line, and the demand forecast of all load nodes; Based on the aforementioned load dynamic electrical characteristics and sensitivity features, the network power flow distribution and node voltage levels under different power output schemes are predicted; With the goal of minimizing network losses and balancing line load rates, the optimal power flow is solved to obtain the optimal operating point and power allocation command for each DC power source.

7. A DC power supply coordination control system based on industrial big data according to claim 6, characterized in that, The process by which the adaptive execution module generates network coordination control commands based on the optimized operating point and power allocation commands includes: Obtain the current upper-level scheduling command or local setting value received by the DC power supply; Calculate the absolute value of the deviation between the optimized operating point and the power allocation command and the current command; Determine whether the absolute value of the deviation is greater than a preset adjustment dead zone threshold; If the deviation exceeds the aforementioned dead zone threshold, a control command sequence is generated based on the deviation. The control commands are issued through the power monitoring network or the energy management system communication protocol. The control objective is to adjust the output power setpoint of the DC power supply or to operate the associated grid-connected switch and power regulation unit.

8. A DC power supply coordination control system based on industrial big data according to claim 7, characterized in that, The adaptive execution module also includes a predictive power quality management unit, which performs the following operations: Based on historical and real-time operational data, the voltage sag, harmonic distortion rate increment, or power shortage that may occur in the distribution network within a preset time period can be predicted. If a predicted event will affect a critical load and the voltage sag, harmonic distortion rate increment, or power shortage exceeds a preset severity threshold, a set of preventative control instructions will be generated in advance. In response to this instruction set, the response strategies of the DC power supply, energy storage system, and reactive power compensation device are coordinated to adjust the network operating status before an event occurs, thereby achieving proactive defense.

Images