A demand-side resource regulation potential dynamic prediction method for power grid resilience

Abstract

The application discloses a demand-side resource regulation potential dynamic prediction method for power grid resilience, and belongs to the technical field of resource regulation, which comprises the following steps: collecting feeder electric signal data, multi-energy flow data, communication quality data and environmental monitoring data; generating a complemented data set by using a neural ordinary differential equation and an integral operator; extracting physical validity features, topological weakening weights, multi-energy output features and environmental disturbance features to generate a comprehensive state matrix; outputting a preliminary prediction value by using a heterogeneous potential prediction model and performing degradation checking by using the topological weakening weights to obtain an effective potential prediction value; calling a rebound constraint model to calculate a surge demand power and eliminating the effective potential prediction value to obtain a potential prediction trajectory. The application deduces feature evolution trajectories by using a neural ordinary differential equation, performs degradation correction by using topological weakening weights, and eliminates the influence of power recovery rebound by using a rebound constraint model, so that the accuracy of effective potential prediction and the recovery safety under extreme disturbance are ensured.

Description

Technical Field

[0001] This invention belongs to the field of resource regulation technology, specifically relating to a dynamic prediction method for demand-side resource regulation potential for power grid resilience. Background Technology

[0002] Demand-side resources for grid resilience refer to various flexible loads, distributed energy facilities, and energy storage devices deployed on the end-user side of the power system and capable of changing their electricity consumption characteristics in response to external dispatch signals. When the power system encounters extreme natural disasters, physical damage, or cyberattacks, these resources enhance the power system's defense capabilities before disturbances occur, its absorption capabilities during disturbances, and its recovery capabilities after disturbances by performing functions such as peak shaving, valley filling, and frequency regulation. Specifically, these resources cover industrial production lines, building air conditioning systems, electric vehicle charging terminals, and user-side photovoltaic systems, exhibiting physical characteristics of wide spatial distribution, rapid dynamic response, and low control costs. Through real-time interaction with the grid dispatch system, demand-side resources for grid resilience can provide load support when the power system is in emergency operation, ensuring the electricity demand of core loads and shortening the time for the power system to resume operation.

[0003] To address the issue of power systems being unable to accurately and in real-time acquire demand-side resource regulation potential under extreme disturbance conditions, existing technologies typically rely on static prediction models based on historical load data and resource assessment algorithms with fixed topologies. However, during processing, issues may arise such as unevenly missing sampling data, neglecting the multi-energy conversion logic of heterogeneous resources, failing to account for the hindering effect of damaged communication links on potential allocation, and lacking quantitative analysis of load rebound power at the moment of power restoration. These issues can lead to a severe disconnect between the potential prediction trajectory and the actual operating conditions, resulting in regulation failures and secondary tripping risks during the power system recovery phase. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, the present invention aims to provide a dynamic prediction method for demand-side resource regulation potential for power grid resilience. This method derives the characteristic evolution trajectory through neural ordinary differential equations, performs degradation correction using topology weakening weights, and eliminates the impact of power restoration rebound using a rebound constraint model, thereby ensuring the accuracy of the actual potential prediction and the safety of recovery under extreme disturbances.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A dynamic prediction method for demand-side resource regulation potential for grid resilience includes the following steps:

[0007] S1. Collect feeder electrical signal data, multi-energy flow data, communication quality data and environmental monitoring data of demand-side resources, and use the constant differential equation and integral operator to deduce and obtain the complete dataset;

[0008] S2. Extract physical effectiveness features, topological weakening weights, multi-energy output features and environmental disturbance features from the completed dataset, perform tensor fusion, and form a comprehensive state matrix.

[0009] S3. Input the integrated state matrix into the heterogeneous potential prediction model using multi-energy hub mapping technology, use the state space equation to map the multi-energy output characteristics to a unified energy density space, adjust the energy density conversion coefficient in the heterogeneous potential prediction model according to the environmental disturbance characteristics, and output the preliminary prediction value.

[0010] S4. Use topological weakening weights to perform degradation verification on the preliminary predicted values ​​to obtain the actual potential predicted values.

[0011] S5. Call the rebound constraint model based on second-order ordinary differential equations and the principle of energy conservation, and calculate the surge in demand power generated at the moment of power restoration by statistically analyzing the energy deficit of demand-side resources during the power outage based on the supplementary dataset.

[0012] S6. Remove the surge in demand power from the actual potential forecast value, obtain the potential forecast trajectory, generate an early warning signal based on the potential forecast trajectory, and adjust the data sampling frequency.

[0013] Preferably, in S1, the process of collecting feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data of demand-side resources includes:

[0014] Power monitoring instruments are deployed at the feeder head nodes of the power system distribution network, multi-energy gateways are deployed at the energy conversion interfaces of demand-side resources, communication quality monitoring units are deployed at the communication routing nodes of demand-side resources, and environmental monitoring sensors are deployed in the area where demand-side resources are located.

[0015] The power monitoring instrument captures the voltage and current waveform signals of the feeder head node at a preset sampling frequency, and uses the fast Fourier transform algorithm to convert the voltage and current waveform signals from the time domain to the frequency domain, extracting the harmonic characteristics and transient voltage fingerprints of specific frequency bands to form feeder electrical signal data.

[0016] The multi-energy gateway periodically reads the register status information of the gas meter, cooling meter, and heating meter through the bus protocol, and converts the original pulse count value and analog signal in the register status information into flow value with physical dimensions to form multi-energy flow data;

[0017] The communication quality monitoring unit sends probe data packets to the communication routing nodes of the demand-side resources, records the sending timestamp and receiving feedback timestamp of the probe data packets, calculates the network latency using the time difference between the sending timestamp and the receiving feedback timestamp, and calculates the packet loss rate based on the number of successfully fed probe data packets per unit time, thus forming communication quality data.

[0018] Environmental monitoring sensors detect physical quantities such as wind force, temperature, and rainfall in the area, convert these physical quantities into continuous analog voltage signals, and perform analog-to-digital conversion and linear calibration on the analog voltage signals to generate environmental monitoring data with physical units.

[0019] Preferably, in S1, the process of obtaining the completed dataset includes:

[0020] The acquired feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data are time-aligned, and the discrete timestamps of each sampling point are labeled. The sampling points corresponding to the discrete timestamps are mapped to the continuous time evolution space using the neural ordinary differential equation to obtain the state latent vector reflecting the resource status on the demand side.

[0021] Construct a state derivative evolution function to map the evolution of the state latent vector over time into a derivative in a continuous time coordinate system;

[0022] For missing data segments in the evolution of the state latent vector over time, the state latent vector at the missing time is obtained by integral derivation of the derivative using the integral operator.

[0023] The latent state vectors obtained from the deduction of missing moments are deconstructed through the output layer of the neural network and restored to the physical dimension values ​​corresponding to the feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data. The restored physical dimension values ​​are then merged with the original sampled data to generate a complete dataset.

[0024] Preferably, in S2, the process of forming the comprehensive state matrix includes:

[0025] Signal decomposition techniques are used to deconstruct the frequency of the feeder electrical signal data in the complete dataset, separating the fundamental component, harmonic characteristics, and transient voltage fingerprint.

[0026] Calculate the phase offset between the fundamental component and the rated frequency signal, and extract the higher harmonic energy distribution from the harmonic characteristics;

[0027] By comparing the transient voltage fingerprint with the pre-stored demand-side resource access feature template, the access integrity of the demand-side resources is identified by calculating the cosine similarity, and the phase offset, higher harmonic energy distribution and access integrity identification results are encapsulated as physical validity features.

[0028] Extract communication quality data from the completed dataset to obtain network latency and packet loss rate;

[0029] Calculate topology weakening weights based on network latency and packet loss rate;

[0030] Extract the feeder electrical signal data from the complete dataset to calculate the active power fluctuation variance, and extract the multi-energy flow data from the complete dataset to calculate the output fluctuation variance of gas flow, cooling flow and heat flow within a preset time window.

[0031] Based on the changes in feeder electrical signal data and multi-energy flow data at continuous sampling times, calculate the average response rate of demand-side resources to external scheduling signals;

[0032] By combining the active power fluctuation variance, output fluctuation variance, and average response rate, a multi-energy output characteristic is constructed to characterize the real-time status of heterogeneous resources such as electricity, gas, cooling, and heating.

[0033] Extract environmental monitoring data from the complete dataset, identify the maximum amplitude of wind speed, temperature and rainfall within the preset observation period, and obtain the peak values ​​of meteorological elements;

[0034] The rate of change of wind speed, temperature and rainfall over time is calculated using the first-order difference operator to obtain the characteristics of environmental evolution trends.

[0035] By combining peak meteorological elements and environmental evolution trend characteristics, environmental disturbance characteristics are constructed.

[0036] Normalization is performed on the physical effectiveness characteristics, topology weakening weights, multi-energy output characteristics, and environmental disturbance characteristics to map them to the same numerical range.

[0037] The normalized physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features are stacked as tensors according to the physical logic dimension and the time dimension to construct a multi-dimensional feature tensor.

[0038] The matrix reshaping operator is used to transform the multidimensional feature tensor into a comprehensive state matrix, where the row vectors of the comprehensive state matrix correspond to the feature dimensions of physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features, and the column vectors correspond to continuous timestamps.

[0039] Preferably, in S3, the process of mapping the multi-energy output characteristics to a unified energy density space using state-space equations includes:

[0040] The integrated state matrix is ​​input into the heterogeneous potential prediction model using multi-energy hub mapping technology, and the feature vector composed of multi-energy output characteristics and the feature vector composed of environmental disturbance characteristics are separated from the integrated state matrix.

[0041] A coupling matrix is ​​constructed based on the physical coupling characteristics of heterogeneous resources such as electricity, gas, cold, and heat. The coupling matrix consists of energy density conversion coefficients that characterize the coupling conversion efficiency of heterogeneous resources. The energy density conversion coefficients serve as matrix elements of the coupling matrix to associate different medium energy sources.

[0042] By using state-space equations to linearly weight and spatially aggregate the output values ​​of each energy dimension in the feature vector composed of multi-energy output features, the multi-energy output features are mapped to a unified energy density space, and energy flow mapping calculation is performed to obtain a preliminary energy flow vector.

[0043] Based on the inverse mapping logic of the coupling matrix and the rated output constraints of demand-side resources, a potential conversion model is constructed to perform dimensional inverse mapping from energy density space to power space.

[0044] Preferably, in S3, the process of outputting the preliminary predicted value includes:

[0045] The initial energy flow vector is adjusted based on the feature vector formed by the characteristics of environmental disturbances;

[0046] The peak values ​​of meteorological elements and the trend characteristics of environmental evolution in the environmental disturbance characteristics are analyzed. The peak values ​​of meteorological elements and the trend characteristics of environmental evolution are input into a preset influence function. The environmental impact correction vector containing the environmental impact correction coefficient is determined through the influence function.

[0047] The energy density conversion coefficients in the coupling matrix are scaled and nonlinearly offset using the environmental impact correction vector to obtain the corrected coupling matrix. The initial energy flow vector is then dynamically compensated using the environmental impact correction vector to obtain the effective energy flow vector mapped to the unified energy density space.

[0048] By combining the modified coupling matrix, the effective energy flow vector is aggregated and constrained in terms of spatiotemporal consistency. The constructed potential conversion model is used to perform dimensional inverse mapping to restore the energy flow density corresponding to the effective energy flow vector to the control power value that matches the modified energy density conversion coefficient, thus obtaining the preliminary prediction value.

[0049] Preferably, in S4, the process of deriving the predicted value of the actual potential includes:

[0050] The analysis of the regulation power values ​​of heterogeneous resources such as electricity, gas, cooling, and heating included in the preliminary forecast values ​​identifies the numerical range of topology weakening weights.

[0051] The initial predicted values ​​are degraded and verified using topological weakening weights to determine the actual potential predicted values.

[0052] The initial prediction value is multiplied by the topology weakening weight to obtain the actual potential prediction value after the communication topology quality degradation.

[0053] When the network latency exceeds the preset time response threshold, it is determined that the demand-side resources cannot complete the response within the time limit required by the external scheduling signal, and the predicted value of the actual potential is corrected to zero.

[0054] When the packet loss rate exceeds the preset communication reliability threshold, it is determined that the communication link between the demand-side resources and the external scheduling center is broken, and the predicted value of the actual potential is corrected to zero.

[0055] Preferably, in S5, the process of calling the bounce constraint model based on second-order ordinary differential equations and the principle of energy conservation includes:

[0056] The rebound constraint model is invoked, and the thermodynamic evolution characteristics of demand-side resources during the regulation process are characterized by second-order ordinary differential equations. State parameters are introduced to quantify the thermodynamic evolution characteristics.

[0057] Based on the physical properties of demand-side resources, a second-order ordinary differential equation is established to describe the evolution of state parameters over time, simulating the dynamic recovery trajectory of demand-side resources after being disturbed.

[0058] Preferably, in S5, the process of calculating the surge in power demand generated at the instant of power restoration includes:

[0059] Identify and complete the feeder power signal data in the dataset. Determine the start time when the active power value in the feeder power signal data is continuously zero as the start time of the power outage, and determine the time when the active power value in the feeder power signal data recovers to a value greater than zero as the time of power restoration.

[0060] During the time interval from the start of the power outage to the time of power restoration, the theoretical power required to keep the state parameters close to zero is calculated using the rebound constraint model.

[0061] Perform time integration on the theoretical power value over the time interval to obtain the energy deficit;

[0062] Based on the principle of energy conservation, the energy conversion relationship between energy deficit and surge demand power is established, and the surge demand power generated at the moment of power restoration is calculated.

[0063] Preferably, in step S6, the process of acquiring the potential prediction trajectory, generating an early warning signal based on the potential prediction trajectory, and adjusting the data sampling frequency includes:

[0064] The surge demand power is removed from the actual potential forecast value by using the subtraction operator to obtain the potential forecast trajectory and the corresponding power value of the potential forecast trajectory.

[0065] The potential prediction trajectory is compared with the preset resilience guarantee threshold in real time. When the power value in the potential prediction trajectory is lower than the resilience guarantee threshold, it is determined that the support capacity of demand-side resources is insufficient to cover the grid resilience guarantee requirements, and an early warning signal is triggered.

[0066] The potential prediction curve is displayed synchronously in the visualization interface. The potential prediction curve includes the rebound range and the effective potential boundary.

[0067] The power smoothing duration in the rebound constraint model is analyzed, and the time period from the power restoration time to the end of the power smoothing duration is marked as the rebound interval;

[0068] The envelope of the power values ​​corresponding to the potential prediction trajectory is marked as the effective potential boundary.

[0069] Adjust the data sampling frequency of power monitoring instruments according to the evolution characteristics of the potential predicted trajectory;

[0070] The power change rate of the potential predicted trajectory is calculated using a first-order difference operator;

[0071] The preset sampling frequency is scaled using the power change rate and the preset frequency adjustment gain constant to obtain the adjusted sampling frequency.

[0072] The adjusted sampling frequency is fed back to the power monitoring instrument to complete the dynamic optimization of the data sampling frequency.

[0073] The beneficial effects of this invention are:

[0074] This invention utilizes the constant differential equation of the God to map discrete sampling points to a continuous-time evolution space, and combines it with integral operators for feature deduction. This solves the problem of non-uniform missing sampling data caused by sensor damage and communication limitations under extreme perturbation environments, ensuring the integrity of the supplementary dataset and the continuity of the prediction process.

[0075] This invention utilizes signal decomposition technology to extract physical validity features from feeder electrical signal data, enabling dynamic identification of the validity of physical access to resources. Combined with multi-energy hub mapping technology and dynamic adjustment of energy density conversion coefficient based on environmental disturbance characteristics, it achieves potential decoupling calculation of multi-energy output features within a unified energy density space, thereby improving the accuracy of preliminary prediction values.

[0076] This invention performs degradation verification on the initial prediction value by introducing a topology weakening weight calculated based on communication quality data, and calls a rebound constraint model based on second-order ordinary differential equations and the principle of energy conservation to quantify and eliminate the surge in demand power generated at the moment of power restoration, thereby generating a potential prediction trajectory that satisfies the physical boundary constraints of the power system. This avoids the safety risks of control command failure caused by inflated actual potential prediction values ​​and secondary tripping during the power system recovery phase. Attached Figure Description

[0077] Figure 1 This is a schematic flowchart of the method of the present invention;

[0078] Figure 2 This is a comparison chart of the completed dataset generated based on the neuron ordinary differential equation during the data resilience verification stage in Embodiment 1 of the present invention.

[0079] Figure 3 This is a simulation comparison diagram of performing degradation verification using topology weakening weights during the topology resilience verification stage in Embodiment 1 of the present invention;

[0080] Figure 4 This is a potential prediction trajectory diagram of the power surge demand quantified and eliminated based on the bounce constraint model during the recovery security verification stage in Embodiment 1 of the present invention.

[0081] Figure 5 This is a simulation result diagram of dynamically adjusting the data sampling frequency based on the potential predicted trajectory evolution characteristics during the adaptive verification stage in Embodiment 1 of the present invention. Detailed Implementation

[0082] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0083] Example 1: As Figure 1 As shown, a dynamic prediction method for demand-side resource regulation potential for grid resilience includes the following steps:

[0084] S1. Collect feeder electrical signal data, multi-energy flow data, communication quality data and environmental monitoring data of demand-side resources, and use the constant differential equation and integral operator to deduce and obtain the complete dataset;

[0085] S2. Extract physical effectiveness features, topological weakening weights, multi-energy output features and environmental disturbance features from the completed dataset, perform tensor fusion, and form a comprehensive state matrix.

[0086] S3. Input the integrated state matrix into the heterogeneous potential prediction model using multi-energy hub mapping technology, use the state space equation to map the multi-energy output characteristics to a unified energy density space, adjust the energy density conversion coefficient in the heterogeneous potential prediction model according to the environmental disturbance characteristics, and output the preliminary prediction value.

[0087] S4. Use topological weakening weights to perform degradation verification on the preliminary predicted values ​​to obtain the actual potential predicted values.

[0088] S5. Call the rebound constraint model based on second-order ordinary differential equations and the principle of energy conservation, and calculate the surge in demand power generated at the moment of power restoration by statistically analyzing the energy deficit of demand-side resources during the power outage based on the supplementary dataset.

[0089] S6. Remove the surge in demand power from the actual potential forecast value, obtain the potential forecast trajectory, generate an early warning signal based on the potential forecast trajectory, and adjust the data sampling frequency.

[0090] The specific implementation process of step S1 is as follows:

[0091] Power monitoring instruments are deployed at the feeder head nodes of the power system distribution network, multi-energy gateways are deployed at the energy conversion interfaces of demand-side resources, communication quality monitoring units are deployed at the communication routing nodes of demand-side resources, and environmental monitoring sensors are deployed in the areas where demand-side resources are located.

[0092] The power monitoring instrument captures the voltage and current waveform signals at the feeder head node at a preset sampling frequency, and uses the fast Fourier transform algorithm to convert the voltage and current waveform signals from the time domain to the frequency domain, extracting the harmonic characteristics and transient voltage fingerprints of specific frequency bands to form feeder electrical signal data.

[0093] The multi-energy gateway periodically reads the register status information of the gas meter, cooling meter, and heating meter through the bus protocol, and converts the original pulse count value and analog signal in the register status information into flow value with physical dimensions to form multi-energy flow data.

[0094] The communication quality monitoring unit sends probe data packets to the communication routing nodes of the demand-side resources, records the timestamp of the probe data packet being sent and the timestamp of the feedback received, calculates the network latency using the time difference between the timestamp of the sent and the timestamp of the feedback received, and calculates the packet loss rate based on the number of successfully fed probe data packets per unit time, thus forming communication quality data.

[0095] Environmental monitoring sensors detect physical quantities such as wind force, temperature, and rainfall in the area, convert these physical quantities into continuous analog voltage signals, and perform analog-to-digital conversion and linear calibration on the analog voltage signals to generate environmental monitoring data with physical units.

[0096] The acquired feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data are time-aligned, and discrete timestamps of each sampling point are labeled. The sampling points corresponding to the discrete timestamps are mapped to the continuous time evolution space using the neural ordinary differential equation to obtain the state latent vector reflecting the resource status on the demand side.

[0097] Constructing the state derivative evolution function , the hidden state vector The evolution over time is mapped to the derivative in a continuous-time coordinate system, where the state implicit vector... The dimension is determined by the state space dimension of the demand-side resources, where t is a continuous timestamp and f is the neural network transformation function. This is the parameter vector of the neural network transformation function.

[0098] For missing data segments in the evolution of the latent state vector over time, the latent state vector at the missing time is obtained by integrating the derivative using an integral operator:

[0099] ;

[0100] in, This represents the latent state vector at the beginning of the missing data segment. The starting time point of the missing data segment. The end time of the missing data segment; the integral term represents the cumulative amount of state changes within the time interval; and the cumulative amount is compared with the latent state vector at the beginning of the missing data segment. The addition enables the deduction of the evolution trajectory of missing features.

[0101] The latent state vector obtained at the missing time point is derived. The neural network output layer is deconstructed to restore the physical dimension values ​​corresponding to the feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data. The restored physical dimension values ​​are then merged with the original sampled data to generate a complete dataset.

[0102] Furthermore, after obtaining the state latent vector for the missing time step by integrating the derivative using the integral operator, and before deconstructing the obtained state latent vector for the missing time step through the output layer of the neural network, a physical validity verification and iterative correction step for the state latent vector can be added. This effectively filters out invalid results in the numerical derivation that do not conform to physical laws, ensuring that the completed dataset maintains both temporal continuity and physical rationality. This improves the accuracy of subsequent comprehensive state matrix construction and potential prediction from the data source, further enhancing the reliability of predictions under extreme perturbations. Specifically, this includes:

[0103] Step 1: Based on the physical laws of the power system, multi-energy system, and environmental monitoring, set reasonable value ranges for each physical quantity, including feeder electrical signal, multi-energy flow, communication quality, and environmental monitoring, as well as the maximum allowable rate of change constraint for each physical quantity over a continuous period of time.

[0104] Step 2: The missing time-state latent vectors obtained from the integral derivation are temporarily deconstructed through the output layer of the neural network to restore the corresponding physical quantity values.

[0105] Step 3: Verify one by one whether the physical dimension values ​​after temporary deconstruction are within the preset reasonable value range, and whether the rate of change of the physical quantity at adjacent time points meets the maximum allowable rate of change constraint.

[0106] Step 4: If there are physical quantities whose values ​​exceed the reasonable range or whose rate of change violates the constraints, calculate the deviation value of the physical quantity, and adaptively fine-tune the parameters of the state derivative evolution function according to the magnitude of the deviation value. Then, use the integral operator to perform integral derivation on the adjusted derivative to obtain the new hidden state vector at the missing time.

[0107] Repeat steps two through four until all physical dimension values ​​after temporary deconstruction meet the preset physical validity constraints. Use the state latent vector that meets the constraints as the final state latent vector at the missing moment, and then perform subsequent deconstruction and merging operations with the original sampled data to generate the complete dataset.

[0108] The specific implementation process of step S2 is as follows:

[0109] Extract physical validity features from the completed dataset.

[0110] Signal decomposition techniques are used to deconstruct the frequency of the feeder electrical signal data in the supplementary dataset, separating the fundamental component, harmonic characteristics, and transient voltage fingerprint.

[0111] Calculate the phase offset between the fundamental component and the rated frequency signal, and extract the higher harmonic energy distribution from the harmonic characteristics.

[0112] By comparing the transient voltage fingerprint with the pre-stored demand-side resource access feature template, the access integrity of the demand-side resources is identified by calculating the cosine similarity, and the phase offset, higher harmonic energy distribution, and access integrity identification results are encapsulated as physical validity features.

[0113] Extract topology weakening weights from the completed dataset.

[0114] Extract communication quality data from the completed dataset to obtain network latency. and packet loss rate .

[0115] Based on network latency and packet loss rate Calculate topology weakening weights ,in, The time delay decay factor, This is the packet loss attenuation factor.

[0116] Extract multi-energy output features from the complete dataset.

[0117] Extract the feeder electrical signal data from the supplementary dataset to calculate the active power fluctuation variance, and extract the multi-energy flow data from the supplementary dataset to calculate the output fluctuation variance of gas flow, cooling flow, and heat flow within a preset time window.

[0118] The average response rate of demand-side resources to external scheduling signals is calculated based on the changes in feeder electrical signal data and multi-energy flow data at continuous sampling times.

[0119] By combining the active power fluctuation variance, output fluctuation variance, and average response rate, a multi-energy output characteristic is constructed to characterize the real-time status of heterogeneous resources such as electricity, gas, cooling, and heat.

[0120] Extract environmental perturbation features from the completed dataset.

[0121] Extract environmental monitoring data from the complete dataset, identify the maximum amplitude of wind speed, temperature and rainfall within the preset observation period, and obtain the peak values ​​of meteorological elements.

[0122] The rate of change of wind speed, temperature, and rainfall over time is calculated using first-order difference operators to obtain the characteristics of environmental evolution trends.

[0123] By combining peak meteorological elements and environmental evolution trend characteristics, environmental disturbance characteristics are constructed.

[0124] Perform tensor fusion to form a comprehensive state matrix.

[0125] The physical effectiveness characteristics, topology weakening weights, multi-energy output characteristics, and environmental disturbance characteristics are normalized to map them to the same numerical range.

[0126] The normalized physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features are stacked as tensors according to the physical logic dimension and the time dimension to construct a multidimensional feature tensor.

[0127] The matrix reshaping operator is used to transform the multidimensional feature tensor into a comprehensive state matrix, where the row vectors of the comprehensive state matrix correspond to the feature dimensions of physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features, and the column vectors correspond to continuous timestamps.

[0128] The specific implementation process of step S3 is as follows:

[0129] The integrated state matrix is ​​input into the heterogeneous potential prediction model using multi-energy hub mapping technology. The feature vector composed of multi-energy output characteristics and the feature vector composed of environmental disturbance characteristics are separated from the integrated state matrix.

[0130] A coupling matrix is ​​constructed based on the physical coupling characteristics of heterogeneous resources such as electricity, gas, cold, and heat. The coupling matrix consists of energy density conversion coefficients that characterize the coupling conversion efficiency of heterogeneous resources. The energy density conversion coefficients serve as matrix elements of the coupling matrix to associate different medium energy sources.

[0131] By using state-space equations to linearly weight and spatially aggregate the output values ​​of each energy dimension in the feature vector composed of multi-energy output characteristics, the multi-energy output characteristics are mapped to a unified energy density space. Energy flow mapping calculations are then performed to obtain a preliminary energy flow vector. Where C is the coupling matrix containing the energy density conversion coefficient, in joules per cubic meter. The feature vector is composed of multi-energy output characteristics.

[0132] Based on the inverse mapping logic of the coupling matrix and the rated output constraints of demand-side resources, a potential conversion model is constructed to perform dimensional inverse mapping from energy density space to power space.

[0133] The initial energy flow vector is adjusted based on the eigenvectors formed by the characteristics of environmental disturbances.

[0134] The peak values ​​of meteorological elements and the trend characteristics of environmental evolution in the environmental disturbance characteristics are analyzed. The peak values ​​of meteorological elements and the trend characteristics of environmental evolution are input into a preset influence function. The influence function is used to determine the environmental impact correction vector, which includes the environmental impact correction coefficient.

[0135] The energy density conversion coefficients in the coupling matrix are scaled and nonlinearly offset using an environmental impact correction vector to obtain a corrected coupling matrix. For the initial energy flow vector Perform dynamic compensation calculations to obtain the effective energy flow vector mapped to a unified energy density space. .

[0136] By combining the modified coupling matrix, the effective energy flow vector is aggregated and constrained in terms of spatiotemporal consistency. The constructed potential conversion model is used to perform dimensional inverse mapping to restore the energy flow density corresponding to the effective energy flow vector to the control power value that matches the modified energy density conversion coefficient, thus obtaining the preliminary prediction value.

[0137] The specific implementation process of step S4 is as follows:

[0138] The analysis of the regulation power values ​​of heterogeneous resources such as electricity, gas, cooling, and heating included in the preliminary forecast values ​​identifies the numerical range of topology weakening weights.

[0139] The initial predicted values ​​are degraded and verified using topological weakening weights to determine the predicted values ​​of actual potential.

[0140] The initial prediction is multiplied by the topology weakening weight to obtain the predicted effective potential value after the communication topology quality degradation. ,in, These are preliminary forecast values. This is used for topological weakening weights.

[0141] When network latency exceeds a preset time response threshold, it is determined that demand-side resources cannot complete the response within the time limit required by the external scheduling signal, and the predicted value of the actual potential is zeroed out.

[0142] When the packet loss rate exceeds the preset communication reliability threshold, it is determined that the communication link between the demand-side resources and the external scheduling center is broken, and the predicted value of the actual potential is corrected to zero.

[0143] The specific implementation process of step S5 is as follows:

[0144] The rebound constraint model is invoked, and second-order ordinary differential equations are used to characterize the thermodynamic evolution characteristics of demand-side resources during the regulation process. State parameters are introduced to quantify the thermodynamic evolution characteristics.

[0145] Establish a second-order ordinary differential equation to describe the evolution of state parameters over time based on the physical properties of demand-side resources. This simulation demonstrates the dynamic recovery trajectory of demand-side resources after a disturbance, where x is a state parameter, M is the equivalent inertia coefficient, D is the equivalent damping coefficient, and K is the equivalent stiffness coefficient. This is the external excitation function.

[0146] Identify and complete the feeder electrical signal data in the dataset. Determine the start time when the active power value in the feeder electrical signal data is continuously zero as the start time of the power outage, and determine the time when the active power value in the feeder electrical signal data recovers to a value greater than zero as the time of power restoration.

[0147] During the time interval from the start of the power outage to the restoration of power, the theoretical power required to maintain the state parameters approaching zero is calculated using the bounce constraint model. .

[0148] For theoretical power values Perform time integration within the time interval to obtain the energy deficit. , The start time of the power outage. This is the moment power was restored.

[0149] Based on the principle of energy conservation, an energy deficit is established. With surging demand for power Energy conversion relationship between Calculate the surge in power demand generated at the moment of power restoration. ,in, The recovery time constant of demand-side resources. This is the preset power smoothing duration.

[0150] The specific implementation process of step S6 is as follows:

[0151] The surge demand power is removed from the actual potential forecast value using a subtraction operator to obtain the potential forecast trajectory and the corresponding power value. ,in, This is a predicted value for actual potential. To meet the surge in power demand.

[0152] The potential prediction trajectory is compared with the preset resilience guarantee threshold in real time. When the power value in the potential prediction trajectory is lower than the resilience guarantee threshold, it is determined that the support capacity of demand-side resources is insufficient to cover the grid resilience guarantee requirements, and an early warning signal is triggered.

[0153] The potential prediction curve is displayed synchronously in the visualization interface. The potential prediction curve includes the rebound range and the effective potential boundary.

[0154] The power stabilization duration in the rebound constraint model is analyzed, and the time period from the power restoration time to the end of the power stabilization duration is marked as the rebound interval.

[0155] The envelope of the power values ​​corresponding to the potential prediction trajectory is marked as the effective potential boundary.

[0156] Adjust the data sampling frequency of power monitoring instruments based on the evolution characteristics of the potential predicted trajectory.

[0157] The power change rate of the potential predicted trajectory is calculated using the first-order difference operator. .

[0158] Using the rate of change of power and preset frequency adjustment gain constant For the preset sampling frequency Perform scaling to obtain the adjusted sampling frequency. ,in, This represents the sampling time interval.

[0159] The adjusted sampling frequency is fed back to the power monitoring instrument to complete the dynamic optimization of the data sampling frequency.

[0160] A specific verification example of the method in this embodiment is as follows:

[0161] Through simulation experiments, the method of this embodiment is verified to achieve accurate prediction of the potential for multi-source heterogeneous demand-side resource regulation and safe control of power grid restoration under complex operating conditions such as non-uniform missing sampling data caused by extreme environmental disturbances, damaged communication links, and load rebound at the moment of power restoration.

[0162] An experimental environment was built using MATLAB, with a distribution network area containing 50 heterogeneous demand-side resources as the simulation object. The resilience event was simulated with T=250s as the power restoration time. At the same time, the non-uniform loss of feeder electrical signal data was simulated in the intervals of 75s-125s and 225s-260s, and the communication link was damaged in the interval of 170s-230s.

[0163] Figure 2 To complete the dataset generation phase, Figure 2 The black dots represent the measured sampling points at the feeder end; the gray dashed line represents the completion result of the existing technology (linear interpolation); and the blue solid line represents the completed dataset generated by the method in this embodiment using the neural network constant differential equation. Figure 2 As can be seen, in the fault intervals corresponding to missing data segments, existing technologies can only perform simple linear connections, resulting in severe distortion of the evolution trajectory. However, the method in this embodiment uses neural ordinary differential equations to map discrete timestamps to continuous time evolution space, restoring the characteristic evolution trajectory that conforms to physical inertia, and verifying the data resilience of the method in this embodiment under extreme perturbations.

[0164] Figure 3 For the degradation verification stage, Figure 3 The pink shaded area represents the identified communication link damage area. Within this area, the gray dashed line represents the initial prediction value, which ignores the impact of communication damage in existing technologies. This value remains high, exhibiting a "falsely high" potential characteristic. The orange solid line represents the effective potential prediction value obtained by the method in this embodiment after degrading the initial prediction value using topology weakening weights. Figure 3 As can be seen, the orange curve actively decreases within the communication link impairment area, accurately depicting the potential transmission obstacles caused by communication obstruction, and verifying the necessity and certainty of the degradation correction method in this embodiment using topology weakening weights.

[0165] Figure 4 To quantify and eliminate surge demand power and potential trajectory generation process in the rebound constraint model. Figure 4 The solid red line represents the surge in demand power generated at the moment of power restoration; the blue shaded area marks the rebound range; the gray dashed line represents the trajectory where existing technologies ignore the impact of load rebound; and the thick green line represents the potential prediction trajectory generated by the method in this embodiment. Figure 4As can be seen, the potential prediction trajectory accurately identifies the physical fact that the control potential is at an extremely low level or in the negative range due to the load rebound at the moment of power restoration after eliminating the surge in demand power. It triggers an early warning signal because it is below the red resilience guarantee threshold. This trend verifies the significant progress of the method in this embodiment in avoiding the risk of secondary tripping during the power system recovery phase.

[0166] Figure 5 This is the stage of dynamic adjustment of data sampling frequency. Figure 5 The purple curve shows the evolution of the adjusted sampling frequency over time; Figure 5 As can be seen, at the moment of power restoration, the potential predicted trajectory power fluctuates violently in a transient state. The method in this embodiment senses the power change rate through a first-order differential operator, so that the sampling frequency is automatically increased in a pulse manner, realizing the dynamic optimization of the sampling frequency of power monitoring instrument data, thereby capturing more high-frequency transient features.

[0167] As can be seen from the above verification process, the method of this embodiment effectively solves the technical pain point of the existing technology in that the potential prediction trajectory under extreme working conditions is out of sync with the actual working conditions by combining the characteristic evolution deduction of the neural ordinary differential equation, the degradation verification considering communication degradation, and the load rebound elimination based on the rebound constraint model. This ensures the accuracy of the actual potential prediction and the safety of recovery under extreme disturbances.

[0168] The dynamic prediction method for demand-side resource regulation potential for power grid resilience proposed in this invention demonstrates excellent data resilience, topological resilience, and recovery security assurance capabilities in the face of extreme disturbance events, and has significant engineering promotion and application value.

[0169] Example 2: A dynamic prediction device for demand-side resource regulation potential for grid resilience, comprising:

[0170] One or more processors;

[0171] Memory, used to store one or more computer programs;

[0172] When one or more programs are executed by one or more processors, the one or more processors execute the method in Example 1.

[0173] Example 3: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method in Example 1.

Claims

1. A dynamic prediction method for demand-side resource regulation potential based on grid resilience, characterized in that, Includes the following steps: S1. Collect feeder electrical signal data, multi-energy flow data, communication quality data and environmental monitoring data of demand-side resources, and use the constant differential equation and integral operator to deduce and obtain the complete dataset; S2. Extract physical effectiveness features, topological weakening weights, multi-energy output features and environmental disturbance features from the completed dataset, perform tensor fusion, and form a comprehensive state matrix. S3. Input the integrated state matrix into the heterogeneous potential prediction model using multi-energy hub mapping technology, use the state space equation to map the multi-energy output characteristics to a unified energy density space, adjust the energy density conversion coefficient in the heterogeneous potential prediction model according to the environmental disturbance characteristics, and output the preliminary prediction value. S4. Use topological weakening weights to perform degradation verification on the preliminary predicted values ​​to obtain the actual potential predicted values. S5. Call the rebound constraint model based on second-order ordinary differential equations and the principle of energy conservation, and calculate the surge in demand power generated at the moment of power restoration by statistically analyzing the energy deficit of demand-side resources during the power outage based on the supplementary dataset. S6. Remove the surge in demand power from the actual potential forecast value, obtain the potential forecast trajectory, generate an early warning signal based on the potential forecast trajectory, and adjust the data sampling frequency.

2. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 1, characterized in that, In S1, the process of collecting feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data from demand-side resources includes: Power monitoring instruments are deployed at the feeder head nodes of the power system distribution network, multi-energy gateways are deployed at the energy conversion interfaces of demand-side resources, communication quality monitoring units are deployed at the communication routing nodes of demand-side resources, and environmental monitoring sensors are deployed in the area where demand-side resources are located. The power monitoring instrument captures the voltage and current waveform signals of the feeder head node at a preset sampling frequency, and uses the fast Fourier transform algorithm to convert the voltage and current waveform signals from the time domain to the frequency domain, extracting the harmonic characteristics and transient voltage fingerprints of specific frequency bands to form feeder electrical signal data. The multi-energy gateway periodically reads the register status information of the gas meter, cooling meter, and heating meter through the bus protocol, and converts the original pulse count value and analog signal in the register status information into flow value with physical dimensions to form multi-energy flow data; The communication quality monitoring unit sends probe data packets to the communication routing nodes of the demand-side resources, records the sending timestamp and receiving feedback timestamp of the probe data packets, calculates the network latency using the time difference between the sending timestamp and the receiving feedback timestamp, and calculates the packet loss rate based on the number of successfully fed probe data packets per unit time, thus forming communication quality data. Environmental monitoring sensors detect physical quantities such as wind force, temperature, and rainfall in the area, convert these physical quantities into continuous analog voltage signals, and perform analog-to-digital conversion and linear calibration on the analog voltage signals to generate environmental monitoring data with physical units.

3. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 1, characterized in that, In S1, the process of obtaining the completed dataset includes: The acquired feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data are time-aligned, and the discrete timestamps of each sampling point are labeled. The sampling points corresponding to the discrete timestamps are mapped to the continuous time evolution space using the neural ordinary differential equation to obtain the state latent vector reflecting the resource status on the demand side. Construct a state derivative evolution function to map the evolution of the state latent vector over time into a derivative in a continuous time coordinate system; For missing data segments in the evolution of the state latent vector over time, the state latent vector at the missing time is obtained by integral derivation of the derivative using the integral operator. The latent state vectors obtained from the deduction of missing moments are deconstructed through the output layer of the neural network and restored to the physical dimension values ​​corresponding to the feeder electrical signal data, multi-energy flow data, communication quality data, and environmental monitoring data. The restored physical dimension values ​​are then merged with the original sampled data to generate a complete dataset.

4. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 1, characterized in that, In S2, the process of forming the comprehensive state matrix includes: Signal decomposition techniques are used to deconstruct the frequency of the feeder electrical signal data in the complete dataset, separating the fundamental component, harmonic characteristics, and transient voltage fingerprint. Calculate the phase offset between the fundamental component and the rated frequency signal, and extract the higher harmonic energy distribution from the harmonic characteristics; By comparing the transient voltage fingerprint with the pre-stored demand-side resource access feature template, the access integrity of the demand-side resources is identified by calculating the cosine similarity, and the phase offset, higher harmonic energy distribution and access integrity identification results are encapsulated as physical validity features. Extract communication quality data from the completed dataset to obtain network latency and packet loss rate; Calculate topology weakening weights based on network latency and packet loss rate; Extract the feeder electrical signal data from the complete dataset to calculate the active power fluctuation variance, and extract the multi-energy flow data from the complete dataset to calculate the output fluctuation variance of gas flow, cooling flow and heat flow within a preset time window. Based on the changes in feeder electrical signal data and multi-energy flow data at continuous sampling times, calculate the average response rate of demand-side resources to external scheduling signals; By combining the active power fluctuation variance, output fluctuation variance, and average response rate, a multi-energy output characteristic is constructed to characterize the real-time status of heterogeneous resources such as electricity, gas, cooling, and heating. Extract environmental monitoring data from the complete dataset, identify the maximum amplitude of wind speed, temperature and rainfall within the preset observation period, and obtain the peak values ​​of meteorological elements; The rate of change of wind speed, temperature and rainfall over time is calculated using the first-order difference operator to obtain the characteristics of environmental evolution trends. By combining peak meteorological elements and environmental evolution trend characteristics, environmental disturbance characteristics are constructed. Normalization is performed on the physical effectiveness characteristics, topology weakening weights, multi-energy output characteristics, and environmental disturbance characteristics to map them to the same numerical range. The normalized physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features are stacked as tensors according to the physical logic dimension and the time dimension to construct a multi-dimensional feature tensor. The matrix reshaping operator is used to transform the multidimensional feature tensor into a comprehensive state matrix, where the row vectors of the comprehensive state matrix correspond to the feature dimensions of physical effectiveness features, topological weakening weights, multi-energy output features, and environmental disturbance features, and the column vectors correspond to continuous timestamps.

5. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 4, characterized in that, In S3, the process of mapping the multi-energy output characteristics to a unified energy density space using state-space equations includes: The integrated state matrix is ​​input into the heterogeneous potential prediction model using multi-energy hub mapping technology, and the feature vector composed of multi-energy output characteristics and the feature vector composed of environmental disturbance characteristics are separated from the integrated state matrix. A coupling matrix is ​​constructed based on the physical coupling characteristics of heterogeneous resources such as electricity, gas, cold, and heat. The coupling matrix consists of energy density conversion coefficients that characterize the coupling conversion efficiency of heterogeneous resources. The energy density conversion coefficients are used as matrix elements of the coupling matrix to associate different medium energy sources. By using state-space equations to linearly weight and spatially aggregate the output values ​​of each energy dimension in the feature vector composed of multi-energy output features, the multi-energy output features are mapped to a unified energy density space, and energy flow mapping calculation is performed to obtain a preliminary energy flow vector. Based on the inverse mapping logic of the coupling matrix and the rated output constraints of demand-side resources, a potential conversion model is constructed to perform dimensional inverse mapping from energy density space to power space.

6. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 5, characterized in that, In S3, the process of outputting the preliminary predicted value includes: The initial energy flow vector is adjusted based on the feature vector formed by the characteristics of environmental disturbances; The peak values ​​of meteorological elements and the trend characteristics of environmental evolution in the environmental disturbance characteristics are analyzed. The peak values ​​of meteorological elements and the trend characteristics of environmental evolution are input into a preset influence function. The influence function is used to determine the environmental impact correction vector, which includes the environmental impact correction coefficient. The energy density conversion coefficients in the coupling matrix are scaled and nonlinearly offset using the environmental impact correction vector to obtain the corrected coupling matrix. The initial energy flow vector is then dynamically compensated using the environmental impact correction vector to obtain the effective energy flow vector mapped to the unified energy density space. By combining the modified coupling matrix, the effective energy flow vector is aggregated and constrained in terms of spatiotemporal consistency. The constructed potential conversion model is used to perform dimensional inverse mapping to restore the energy flow density corresponding to the effective energy flow vector to the control power value that matches the modified energy density conversion coefficient, thus obtaining the preliminary prediction value.

7. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 6, characterized in that, In S4, the process of deriving the predicted effective potential includes: The analysis of the regulation power values ​​of heterogeneous resources such as electricity, gas, cooling, and heating included in the preliminary forecast values ​​identifies the numerical range of topology weakening weights. The initial predicted values ​​are degraded and verified using topological weakening weights to determine the actual potential predicted values. The initial prediction value is multiplied by the topology weakening weight to obtain the actual potential prediction value after the communication topology quality degradation. When the network latency exceeds the preset time response threshold, it is determined that the demand-side resources cannot complete the response within the time limit required by the external scheduling signal, and the predicted value of the actual potential is corrected to zero. When the packet loss rate exceeds the preset communication reliability threshold, it is determined that the communication link between the demand-side resources and the external scheduling center is broken, and the predicted value of the actual potential is corrected to zero.

8. The method for dynamic prediction of demand-side resource regulation potential for grid resilience according to claim 1, characterized in that, In S5, the process of calling the bounce constraint model based on second-order ordinary differential equations and the energy conservation principle includes: The rebound constraint model is invoked, and the thermodynamic evolution characteristics of demand-side resources during the regulation process are characterized by second-order ordinary differential equations. State parameters are introduced to quantify the thermodynamic evolution characteristics. Based on the physical properties of demand-side resources, a second-order ordinary differential equation is established to describe the evolution of state parameters over time, simulating the dynamic recovery trajectory of demand-side resources after being disturbed.

9. A dynamic prediction method for demand-side resource regulation potential for grid resilience according to claim 8, characterized in that, In S5, the process of calculating the surge in power demand generated at the moment of power restoration includes: Identify and complete the feeder power signal data in the dataset. Determine the start time when the active power value in the feeder power signal data is continuously zero as the start time of the power outage, and determine the time when the active power value in the feeder power signal data recovers to a value greater than zero as the time of power restoration. During the time interval from the start of the power outage to the time of power restoration, the theoretical power required to keep the state parameters close to zero is calculated using the rebound constraint model. Perform time integration on the theoretical power value over the time interval to obtain the energy deficit; Based on the principle of energy conservation, the energy conversion relationship between energy deficit and surge demand power is established, and the surge demand power generated at the moment of power restoration is calculated.

10. A dynamic prediction method for demand-side resource regulation potential for grid resilience according to claim 1, characterized in that, In S6, the process of acquiring the potential prediction trajectory, generating an early warning signal based on the potential prediction trajectory, and adjusting the data sampling frequency includes: The surge demand power is removed from the actual potential forecast value by using the subtraction operator to obtain the potential forecast trajectory and the corresponding power value of the potential forecast trajectory. The potential prediction trajectory is compared with the preset resilience guarantee threshold in real time. When the power value in the potential prediction trajectory is lower than the resilience guarantee threshold, it is determined that the support capacity of demand-side resources is insufficient to cover the grid resilience guarantee requirements, and an early warning signal is triggered. The potential prediction curve is displayed synchronously in the visualization interface. The potential prediction curve includes the rebound range and the effective potential boundary. The power smoothing duration in the rebound constraint model is analyzed, and the time period from the power restoration time to the end of the power smoothing duration is marked as the rebound interval; The envelope of the power values ​​corresponding to the potential prediction trajectory is marked as the effective potential boundary. Adjust the data sampling frequency of power monitoring instruments according to the evolution characteristics of the potential predicted trajectory; The power change rate of the potential predicted trajectory is calculated using a first-order difference operator; The preset sampling frequency is scaled using the power change rate and the preset frequency adjustment gain constant to obtain the adjusted sampling frequency. The adjusted sampling frequency is fed back to the power monitoring instrument to complete the dynamic optimization of the data sampling frequency.

Images