Intelligent Control Method and System for Power Supply in Oil Well Fracturing Based on Supercapacitors

By constructing a nonlinear dynamic model and using online identification technology, the problem of insufficient power demand prediction in the oil well fracturing power supply system was solved, achieving precise power supply and fault early warning, and improving the system's operational reliability and efficiency.

CN123121691APending Publication Date: 2026-11-17BEIJING RUIHE DEBAO THERMAL TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611351735.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-02
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing oil well fracturing power supply systems cannot anticipate power demand when faced with nonlinear and time-varying load characteristics, resulting in frequent deep charging and discharging of supercapacitors, unstable state of charge, and impact on power supply reliability and supercapacitor lifespan.

Method used

By collecting multi-source heterogeneous data, performing multi-scale time-frequency decomposition, constructing a nonlinear dynamic model, predicting future power demand, establishing a dynamic mapping relationship between the state of charge of supercapacitors and fracturing conditions, generating a globally optimal energy dispatch trajectory, and updating the equivalent circuit model parameters through online identification technology, we can achieve precise power supply and fault early warning.

Benefits of technology

It improves the stability of power supply voltage and frequency quality, extends the service life of supercapacitor banks, reduces the probability of unplanned downtime and maintenance costs, and enhances the operational reliability and energy utilization efficiency of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121691A_ABST
    Figure CN123121691A_ABST
Patent Text Reader

Abstract

This invention relates to the field of oil well fracturing power supply technology, and particularly to an intelligent control method and system for oil well fracturing power supply based on supercapacitors. The method includes: collecting multi-source heterogeneous data from the fracturing site; extracting transient impact and steady-state energy consumption characteristics through multi-scale time-frequency decomposition; constructing a load nonlinear dynamic model to predict power demand across multiple time windows; establishing a dynamic mapping between the supercapacitor's state of charge and fracturing conditions; solving a multi-objective optimization problem with constraints on energy, power, and lifetime to generate a globally optimal energy scheduling trajectory and perform rolling corrections; coordinating power supply between the supercapacitor bank and the main power source through adaptive power allocation control; online identification and updating of capacitor equivalent model parameters; generating joint decisions on fault warning, topology reconstruction, and scheduling compensation when drift exceeds a threshold; feeding these decisions back to the nonlinear dynamic model to form a closed-loop optimization, thereby achieving intelligent optimization and reliable operation of fracturing power supply.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of oil well fracturing power supply technology, and in particular to an intelligent control method and system for oil well fracturing power supply based on supercapacitors. Background Technology

[0002] Oil well fracturing is a crucial step in enhancing oil and gas field production. Fracturing pump trucks generate drastically fluctuating impact loads during operation, especially during pump start-up, shutdown, and gear shifting, where instantaneous power surges can reach several times the steady-state power. Currently, the power supply system at fracturing sites typically uses diesel generators as the main power source, supplemented by supercapacitor energy storage devices to absorb or compensate for power deficits. For this hybrid power supply structure, existing control methods mostly employ power allocation strategies based on threshold judgment or empirical rules. This involves setting fixed voltage or power thresholds; when the load power exceeds the threshold, the supercapacitor discharges to support the load; when it falls below the threshold, it switches to charging mode. Some engineering applications also introduce conventional PID control or simple logic switching methods to achieve energy coordination between the supercapacitor and the main power source.

[0003] Fracturing operations are characterized by strong nonlinearity and time-varying properties. Load characteristics are closely related to well conditions, sand ratio, pump pressure, and other factors. Existing conventional control methods sever the connection between historical operating conditions and future load trends, allowing only passive responses to current transient power and failing to proactively predict the evolution of power demand. This leads to frequent deep charging and discharging of supercapacitors, causing their state of charge to easily fall into excessively low or high ranges, weakening peak shaving and valley filling effects and accelerating capacitor aging. The equivalent circuit parameters of supercapacitors drift significantly with temperature and the number of charge-discharge cycles. Conventional methods typically implement control based on a fixed factory model, ignoring control errors caused by parameter drift. When energy storage capacity declines, timely correction of scheduling behavior is impossible, making it difficult to guarantee power supply reliability. Summary of the Invention

[0004] This invention provides an intelligent control method and system for power supply in oil well fracturing based on supercapacitors, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides an intelligent control method for power supply in oil well fracturing based on supercapacitors, comprising:

[0006] Multi-source heterogeneous data from fracturing operations are collected, and multi-scale time-frequency decomposition is performed on the multi-source heterogeneous data to extract the transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of fracturing load is constructed, and the evolution trajectory of power demand in multiple future time windows is predicted based on the nonlinear dynamic model.

[0007] Based on the power demand evolution trajectory, a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence is established. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, a globally optimal energy scheduling trajectory covering pre-charging timing, discharge depth allocation and main power supply coordination ratio is generated, and the globally optimal energy scheduling trajectory is rolled over during execution.

[0008] Based on the global optimal energy scheduling trajectory, the energy interaction between the supercapacitor bank and the main power supply is controlled by the adaptive power distribution unit to achieve precise power supply to the fracturing equipment;

[0009] The equivalent circuit model parameters of the supercapacitor bank are updated in real time by online identification technology. When the drift of the model parameters exceeds the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

[0010] Multi-scale time-frequency decomposition is performed on the multi-source heterogeneous data to extract transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of fracturing load is constructed, and based on the nonlinear dynamic model, the power demand evolution trajectory for multiple future time windows is predicted, including:

[0011] Time alignment and sampling rate normalization are performed on multi-source heterogeneous data to construct a multi-dimensional data matrix under a unified time reference.

[0012] Short-time Fourier transform and wavelet packet decomposition are performed on the multidimensional data matrix to obtain time-frequency representations at different scales. The amplitude envelope and duration of transient power impact events are identified from the time-frequency representations. The frequency domain dominant components of the steady-state energy consumption stage are extracted. The transient power impact features and steady-state energy consumption features are fused and mapped to generate the operating condition feature vector.

[0013] Based on the aforementioned operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted by nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships.

[0014] The current working condition feature vector is used as the initial state of the nonlinear dynamic model. The displacement sequence and sand ratio sequence in the fracturing construction plan are combined as future driving inputs. The state evolution equation of the nonlinear dynamic model in multiple future time windows is solved by numerical integration method to obtain the power demand evolution trajectory including power amplitude, rate of change and duration.

[0015] Based on the aforementioned operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted using nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships.

[0016] The working condition feature vector is subjected to a delay embedding transformation. The time delay parameter is determined by calculating the mutual information function, and the embedding dimension parameter is determined by calculating the correlation dimension. The one-dimensional working condition feature vector sequence is reconstructed into a set of trajectory points in a high-dimensional phase space using the time delay parameter and the embedding dimension parameter.

[0017] A topological invariant analysis is performed on the trajectory point set to extract the Lyapunov exponent spectrum and attractor geometric features of the trajectory point set. The chaotic characteristics and stability domain boundary of the system are determined based on the Lyapunov exponent spectrum, and the state variable dimension and coupling form of the fracturing load dynamic system are determined based on the attractor geometric features, thereby generating a structural framework for the state space expression.

[0018] The state equations and observation equations in the structural framework of the state-space expression are parameterized. The state equations describe the time evolution of state variables, and the observation equations describe the mapping relationship between state variables and measurable power demand. Fracturing construction parameters are introduced into the state equations as driving inputs.

[0019] A recursive identification strategy is used to estimate the undetermined parameters in the state equation and the observation equation online. The parameter values ​​are updated iteratively by minimizing the cost function of the prediction error. A time-varying weighting factor is introduced to attenuate the historical observation data to track the changes in operating conditions, resulting in a nonlinear dynamic model that includes the evolution of state variables, driving input response and coupling between states.

[0020] By solving a multi-objective optimization problem that includes energy constraints, power constraints, and lifetime constraints, a globally optimal energy scheduling trajectory is generated, encompassing pre-charge timing, discharge depth allocation, and main power supply coordination ratio.

[0021] Based on the power demand evolution trajectory, the power fluctuation amplitude sequence and duration sequence of fracturing operations are extracted, the releaseable energy of the supercapacitor bank at different discharge depths and the corresponding cycle life loss rate are calculated, and a nonlinear mapping relationship between discharge depth and cycle life loss rate is constructed.

[0022] The nonlinear mapping relationship is used as a mathematical representation of the lifetime constraint. An energy constraint inequality is established by combining the upper and lower limits of the state of charge of the supercapacitor bank. A power constraint inequality is established by combining the physical limits of charging and discharging power. The peak load suppression target of the main power supply and the energy cycle efficiency target of the supercapacitor bank are introduced to construct a multi-objective optimization problem.

[0023] The multi-objective optimization problem is solved in layers. In the outer layer optimization, the trade-off coefficient between energy utilization and lifetime loss is determined by Pareto front search. In the inner layer optimization, the optimal decision sequence that satisfies all constraints is solved based on the trade-off coefficient.

[0024] In the inner optimization, the decision variables are discretized into a set of pre-charge start-up times, a discharge depth allocation vector for each time period, and a main power supply coordination ratio vector. A dynamic programming strategy is used to solve the problem recursively for each time period. The final value of the state of charge of the previous time period is used as the initial value of the state transition of the current time period. The impact of the decision of the current time period on the feasible region of the subsequent time periods is calculated through the state transition equation. Finally, the global optimal energy scheduling trajectory is obtained by backtracking.

[0025] Using the aforementioned nonlinear mapping relationship as a mathematical representation of lifetime constraints, energy constraint inequalities are established in conjunction with the upper and lower limits of the supercapacitor's state of charge, and power constraint inequalities are established in conjunction with the physical limits of charge and discharge power. Furthermore, by introducing the main power supply peak load suppression objective and the supercapacitor's energy cycle efficiency objective, a multi-objective optimization problem is constructed, including:

[0026] The discharge depth and cycle lifetime loss rate in the nonlinear mapping relationship are transformed into a cumulative lifetime loss expression. The lifetime loss at each moment in the entire scheduling cycle is accumulated, and a lifetime constraint inequality is established to limit the cumulative lifetime loss to not exceed the health threshold.

[0027] Based on the cumulative lifetime loss expression, the contribution weight of the discharge depth to the total lifetime loss at each moment is calculated, the lifetime loss sensitive period is identified, the upper and lower limits of the supercapacitor bank's state of charge are converted into the energy reserve requirements and energy retention requirements of the lifetime loss sensitive period, and an energy constraint inequality is established.

[0028] Extract the dynamic boundary of the physical limit of charging and discharging power with respect to the state of charge, and establish a power constraint inequality to constrain the charging and discharging power at each moment to be within the feasible region determined by the dynamic boundary.

[0029] The peak load suppression objective of the main power supply is defined as the expression for minimizing the variance of the power supply power fluctuation, and the energy cycle efficiency objective of the supercapacitor bank is defined as the expression for minimizing the charging and discharging energy loss. The peak load suppression objective of the main power supply and the energy cycle efficiency objective of the supercapacitor bank are assigned weight coefficients and weighted and summed to form a comprehensive objective. The comprehensive objective is combined with the lifetime constraint inequality, the energy constraint inequality and the power constraint inequality to construct a multi-objective optimization problem.

[0030] The equivalent circuit model parameters of the supercapacitor bank are updated in real time using online identification technology. When the drift of the model parameters exceeds the adaptive threshold, a joint decision-making scheme is generated, which includes fault warning, topology reconfiguration instructions, and scheduling strategy compensation.

[0031] The terminal voltage and current responses of the supercapacitor bank are collected, and the equivalent series resistance, equivalent parallel capacitance and leakage resistance parameters in the equivalent circuit model are calculated through online identification technology. The calculated parameter values ​​are used as the parameters of the equivalent circuit model updated in real time.

[0032] Calculate the parameter drift of the real-time updated equivalent circuit model parameters relative to the initial calibration value, extract the drift rate and drift acceleration based on the time series features of the parameter drift, and dynamically generate an adaptive threshold according to the coupling relationship between the drift rate and the drift acceleration;

[0033] When the parameter drift exceeds the adaptive threshold, the contribution ratio of each model parameter in the parameter drift is analyzed, the dominant degradation parameter is identified, and fault warning information is generated based on the physical meaning of the dominant degradation parameter.

[0034] Based on the fault warning information, the location of the degraded individual cell in the supercapacitor bank is located, a topology reconfiguration instruction is generated to isolate the individual cell location and reconfigure the electrical connection path, the power output capacity attenuation coefficient and the energy reserve attenuation coefficient are calculated based on the equivalent parameters of the reconfigured topology, and a scheduling strategy compensation scheme is generated based on the power output capacity attenuation coefficient and the energy reserve attenuation coefficient.

[0035] The fault warning information, the topology reconfiguration instruction, and the scheduling strategy compensation scheme are combined to form a joint decision-making scheme.

[0036] The equivalent series resistance, equivalent parallel capacitance, and leakage resistance parameters in the equivalent circuit model are calculated using online identification technology. The calculated parameter values ​​are then used as the parameters of the equivalent circuit model in real time.

[0037] The amplitude of the terminal voltage jump and the current step response of the supercapacitor bank during the charging and discharging transition are collected, and the current resistance value of the equivalent series resistance is calculated based on the ratio of the amplitude of the terminal voltage jump to the current step response.

[0038] The terminal voltage rise curve of the supercapacitor bank during the constant current charging stage is collected. The voltage change rate sequence is obtained by performing time-domain differentiation on the terminal voltage rise curve. The real-time capacitance value of the equivalent parallel capacitor is inverted based on the numerical relationship between the voltage change rate sequence and the charging current.

[0039] The self-discharge decay curve of the terminal voltage of the supercapacitor bank during the resting stage is collected. The decay time constant is extracted by exponential fitting of the terminal voltage self-discharge decay curve. The current value of the leakage resistance is calculated based on the decay time constant and the real-time capacitance value.

[0040] The current resistance value of the equivalent series resistor, the real-time capacitance value of the equivalent parallel capacitor, and the current resistance value of the leakage resistor are timestamped to establish a parameter update record sequence. The parameter values ​​of multiple consecutive sampling periods in the parameter update record sequence are smoothed by a sliding window to eliminate the interference of transient noise on parameter identification. The filtered parameter values ​​are used as the parameters of the equivalent circuit model after real-time update.

[0041] A second aspect of the present invention provides an intelligent control system for power supply to oil well fracturing based on supercapacitors, comprising:

[0042] The acquisition and prediction unit is used to acquire multi-source heterogeneous data from the fracturing operation site, perform multi-scale time-frequency decomposition on the multi-source heterogeneous data, extract the transient power impact characteristics and steady-state energy consumption characteristics of the fracturing operation, construct a nonlinear dynamic model of the fracturing load, and predict the power demand evolution trajectory of multiple future time windows based on the nonlinear dynamic model.

[0043] The energy scheduling unit is used to establish a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence based on the power demand evolution trajectory. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, it generates a globally optimal energy scheduling trajectory that covers pre-charging timing, discharge depth allocation and main power supply coordination ratio. During the execution process, the globally optimal energy scheduling trajectory is rolled over and corrected.

[0044] The power distribution unit is used to control the energy interaction between the supercapacitor bank and the main power supply through the adaptive power distribution unit according to the global optimal energy scheduling trajectory, so as to achieve precise power supply to the fracturing equipment.

[0045] The identification feedback unit is used to update the equivalent circuit model parameters of the supercapacitor bank in real time through online identification technology. When the drift of the model parameters is detected to exceed the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

[0046] A third aspect of the present invention provides an electronic device, comprising:

[0047] processor;

[0048] Memory used to store processor-executable instructions;

[0049] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0050] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0051] To address the characteristics of fracturing operations—namely, the strong impact, nonlinearity, and multi-timescale coupling of loads—multi-scale time-frequency decomposition of multi-source heterogeneous data effectively separates transient power surges from steady-state energy consumption components. This enables the constructed nonlinear dynamic model to accurately characterize complex operating conditions. Based on this model, the evolution trajectory of future multi-window power demand is predicted, significantly improving the accuracy of predicting transient surge peak values ​​and durations. This provides a reliable basis for proactive control, avoiding overload trips or power supply redundancy caused by power prediction errors.

[0052] A dynamic mapping relationship between the state of charge of supercapacitors and the fracturing operation sequence is established, and a multi-objective optimization problem encompassing energy, power, and lifetime constraints is solved. This enables dynamic coordination of the output ratio between the supercapacitor bank and the main power supply at different fracturing stages, obtaining the globally optimal energy dispatch trajectory. A rolling correction mechanism during execution allows the dispatch strategy to respond in real time to sudden changes in operating conditions, effectively suppressing load shocks that disrupt the power grid, reducing peak power demand from the main power supply, extending the cycle life of the supercapacitor bank, and improving the overall energy utilization efficiency of the system.

[0053] Based on the globally optimal energy scheduling trajectory, the energy interaction between the supercapacitor bank and the main power supply is controlled by the adaptive power distribution unit, achieving millisecond-level dynamic power compensation and precise power supply, which greatly improves the stability of the power supply voltage and frequency quality, ensuring the continuous and stable operation of the fracturing equipment under harsh working conditions.

[0054] Online identification technology updates the parameters of the supercapacitor's equivalent circuit model in real time, enabling timely detection of aging signs such as capacitance decay and increased internal resistance. When parameter drift exceeds limits, a joint decision-making scheme integrating fault warning, topology reconfiguration, and scheduling strategy compensation takes effect quickly, preventing fault propagation. The execution results are fed back to the nonlinear dynamic model to form a closed-loop optimization, giving the system self-learning and adaptive capabilities, continuously improving operational reliability throughout its entire lifecycle, and significantly reducing the probability of unplanned downtime and maintenance costs. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the intelligent control method for power supply to oil well fracturing based on supercapacitors, according to an embodiment of the present invention.

[0056] Figure 2 This is an image showing the extraction of the transient power impulse amplitude envelope and duration based on short-time Fourier transform and wavelet packet decomposition in an embodiment of the present invention.

[0057] Figure 3This is a global optimal energy dispatch trajectory diagram of the supercapacitor and main power supply working together over multiple time periods in an embodiment of the present invention;

[0058] Figure 4 This is a flowchart of the supercapacitor bank energy management and fault early warning control according to an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0061] Figure 1 This is a flowchart illustrating the intelligent control method for oil well fracturing power supply based on supercapacitors, according to an embodiment of the present invention.

[0062] The intelligent control method for power supply in oil well fracturing based on supercapacitors includes:

[0063] Multi-source heterogeneous data from fracturing operations are collected, and multi-scale time-frequency decomposition is performed on the multi-source heterogeneous data to extract the transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of fracturing load is constructed, and the evolution trajectory of power demand in multiple future time windows is predicted based on the nonlinear dynamic model.

[0064] Based on the power demand evolution trajectory, a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence is established. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, a globally optimal energy scheduling trajectory covering pre-charging timing, discharge depth allocation and main power supply coordination ratio is generated, and the globally optimal energy scheduling trajectory is rolled over during execution.

[0065] Based on the global optimal energy scheduling trajectory, the energy interaction between the supercapacitor bank and the main power supply is controlled by the adaptive power distribution unit to achieve precise power supply to the fracturing equipment;

[0066] The equivalent circuit model parameters of the supercapacitor bank are updated in real time by online identification technology. When the drift of the model parameters exceeds the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

[0067] In one optional implementation, the multi-source heterogeneous data is decomposed into multi-scale time-frequency components to extract transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of the fracturing load is constructed, and the power demand evolution trajectory for multiple future time windows is predicted based on the nonlinear dynamic model, including:

[0068] Time alignment and sampling rate normalization are performed on multi-source heterogeneous data to construct a multi-dimensional data matrix under a unified time reference.

[0069] Short-time Fourier transform and wavelet packet decomposition are performed on the multidimensional data matrix to obtain time-frequency representations at different scales. The amplitude envelope and duration of transient power impact events are identified from the time-frequency representations. The frequency domain dominant components of the steady-state energy consumption stage are extracted. The transient power impact features and steady-state energy consumption features are fused and mapped to generate the operating condition feature vector.

[0070] Based on the aforementioned operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted by nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships.

[0071] The current working condition feature vector is used as the initial state of the nonlinear dynamic model. The displacement sequence and sand ratio sequence in the fracturing construction plan are combined as future driving inputs. The state evolution equation of the nonlinear dynamic model in multiple future time windows is solved by numerical integration method to obtain the power demand evolution trajectory including power amplitude, rate of change and duration.

[0072] like Figure 2 As shown, the method includes:

[0073] This figure illustrates the transient feature extraction results of the fracturing load signal after multi-scale time-frequency decomposition. The horizontal axis represents time (seconds), and the vertical axis represents the normalized amplitude. The spike pulses in the curve correspond to the power impacts generated during the start-up, shutdown, and gear shifting stages of the fracturing pump. The time-frequency spectrum is obtained through short-time Fourier transform (64 ms window), and the power envelope curve is obtained by integration along the frequency axis. The adaptive threshold (dashed line in the figure) is dynamically set based on the root mean square of the envelope. When the envelope exceeds the threshold, it is marked as the start of the impact; when it falls back below the threshold, it is marked as the end of the impact, thus accurately extracting the peak amplitude and duration of each impact (e.g., peak value 2 at 0.52 s, lasting 0.04 s). Wavelet packet decomposition (4 layers) further separates high-frequency details from low-frequency steady-state components, ensuring that the transient features are not affected by background noise. The extracted amplitude envelope and duration sequence, as important components of the operating condition feature vector, provide accurate transient excitation input for subsequent nonlinear dynamic modeling, supporting load prediction and energy scheduling decisions.

[0074] At fracturing operations, the collected data comes from a wide range of heterogeneous sources, including instantaneous power sensors from fracturing pump units, hydraulic pressure transmitters, flow meters, online sand ratio analyzers, and voltage and current transformers from the main power supply. These sensors exhibit significant differences in sampling frequencies and inconsistent timestamp alignment accuracy. To address this issue, time alignment processing was first performed on all data channels. Using the highest global sampling frequency as the reference time axis, cubic spline interpolation was used for upsampling of low-frequency data channels, while anti-aliasing low-pass filtering was applied to high-frequency data channels for downsampling. This unified all sensor data to the same time resolution. After sampling rate normalization, the data channels were arranged along the time axis to construct a multi-dimensional data matrix. ,in Indicates the total number of data channels. This represents the total number of sampling points under a unified time reference. This matrix provides a structurally regular and time-referenced input for subsequent multi-scale time-frequency decomposition.

[0075] For multidimensional data matrices Each data channel in the process undergoes a short-time Fourier transform and wavelet packet decomposition, with each method having its own emphasis. The short-time Fourier transform uses a fixed window length. The signal is segmented into frames and windowed, and a Discrete Fourier Transform is performed on each frame to obtain a two-dimensional time-frequency distribution. This effectively captures the frequency composition of the signal at different times, making it particularly suitable for analyzing events with local frequency abrupt changes, such as the start-up and shutdown of fracturing pumps. Wavelet packet decomposition uses a multi-scale filter bank to recursively decompose the signal, preserving the dominant component of steady-state energy accumulation in the low-frequency subband and highlighting the detailed features of transient impacts in the high-frequency subband. The decomposition level is [number missing]. The frequency is determined based on the ratio of the lowest frequency of interest to the sampling frequency, and is usually taken as... The number of layers ranges from 4 to 6. The results of the two transformations complement each other, together forming a set of time-frequency representations at different scales.

[0076] When identifying transient power surge events from time-frequency representations, the time spectrum of the short-time Fourier transform is integrated along the frequency axis to obtain the envelope curve of power changing with time, and an adaptive amplitude threshold is set. When the envelope curve exceeds at a certain moment The event is considered to have started when the envelope curve falls back to the initial value. The following conditions are considered the end of the impact event, and the peak amplitude of each impact event is extracted accordingly. With duration For the steady-state energy consumption stage, the center frequency of the dominant frequency component is identified by the energy spectrum of the low-frequency subband coefficients obtained from wavelet packet decomposition. and its energy percentage These two parameters reflect the operating frequency characteristics and energy distribution patterns of the fracturing pump unit during the steady-state phase. Transient power impact characteristics (including...) , ) and steady-state energy consumption characteristics (including , The data, along with current process parameters such as displacement and sand ratio, are combined and normalized to form a feature vector of the operating condition. ,in This represents the total number of feature dimensions. This fusion mapping process compresses multi-dimensional information from the time domain, frequency domain, and process domain into a unified feature space, providing a high-quality input representation for subsequent dynamic modeling.

[0077] Based on the characteristic vector sequence of operating conditions, a state-space reconstruction method is used to establish the phase space trajectory of fracturing load. Specifically, the embedding dimension is used... and time delay Using parameters, a delayed embedding is performed on the time series of historical operating condition feature vectors, mapping one-dimensional or low-dimensional time series to a high-dimensional phase space, allowing the intrinsic dynamic structure of the fracturing load system to be fully unfolded. Embedding dimension Time delay is estimated using the pseudo-nearest neighbor method. The first minimum value of the mutual information function is used to determine the phase space. In the reconstructed phase space, historical data form a trajectory whose geometry reflects the attractor structure of the fracturing load system. The evolution of the phase space trajectory is fitted using nonlinear system identification techniques, specifically a hybrid identification framework combining Gaussian process regression and neural ordinary differential equations: Gaussian process regression captures local nonlinear mappings in the trajectory evolution, providing uncertainty estimates; neural ordinary differential equations parameterize the state transition function in the form of a continuous-time dynamic system, enabling the model to perform integral derivations at arbitrary step sizes. The final nonlinear dynamic model includes state variables. (Corresponding to the current state point in the reconstructed phase space), driving input (Controllable process parameters such as displacement sequence and sand ratio sequence) and the coupling relationship function between state variables and drive input. The model is in the form of ,in It represents the derivative of the state variable with respect to time.

[0078] In the prediction phase of power demand evolution trajectory, the current operating condition characteristic vector is used as the initial state of the nonlinear dynamic model. Extract the displacement sequence for each future time step from the fracturing operation plan. sand ratio sequence ( (Indexed by future time steps), combining them into a future-driven input sequence. The fourth-order Runge-Kutta numerical integration method is used to solve the state evolution equations of the nonlinear dynamic model over multiple time windows, progressively advancing the state variables. The evolution of the state variables. At the end of each integration step, the current state variable is decoded into the power demand prediction value at the corresponding time moment, including the power magnitude, through the output mapping function. Power change rate and the duration of this power level Arrange the forecast results for all time windows in chronological order to obtain the power demand evolution trajectory. This trajectory fully describes the dynamic changes in power demand during future fracturing operations, providing precise feedforward input for subsequent energy scheduling optimization of the supercapacitor bank.

[0079] In practical engineering applications, fracturing operation plans are often subject to temporary adjustments, causing changes in the input sequence during the operation. To address this, a sliding window mechanism is introduced into the prediction process, where the latest feature vector of the operating conditions is re-acquired at fixed time intervals to update the initial state. The system updates the driving input sequence with the latest construction plan parameters, re-executes numerical integration, and refreshes the power demand evolution trajectory. This rolling prediction mechanism enables the model to respond promptly to real-time changes in on-site conditions, maintaining the timeliness and accuracy of prediction results, while also providing continuously updated prediction data for subsequent rolling corrections of the energy dispatch trajectory.

[0080] In one optional implementation, based on the operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted using nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships.

[0081] The working condition feature vector is subjected to a delay embedding transformation. The time delay parameter is determined by calculating the mutual information function, and the embedding dimension parameter is determined by calculating the correlation dimension. The one-dimensional working condition feature vector sequence is reconstructed into a set of trajectory points in a high-dimensional phase space using the time delay parameter and the embedding dimension parameter.

[0082] A topological invariant analysis is performed on the trajectory point set to extract the Lyapunov exponent spectrum and attractor geometric features of the trajectory point set. The chaotic characteristics and stability domain boundary of the system are determined based on the Lyapunov exponent spectrum, and the state variable dimension and coupling form of the fracturing load dynamic system are determined based on the attractor geometric features, thereby generating a structural framework for the state space expression.

[0083] The state equations and observation equations in the structural framework of the state-space expression are parameterized. The state equations describe the time evolution of state variables, and the observation equations describe the mapping relationship between state variables and measurable power demand. Fracturing construction parameters are introduced into the state equations as driving inputs.

[0084] A recursive identification strategy is used to estimate the undetermined parameters in the state equation and the observation equation online. The parameter values ​​are updated iteratively by minimizing the cost function of the prediction error. A time-varying weighting factor is introduced to attenuate the historical observation data to track the changes in operating conditions, resulting in a nonlinear dynamic model that includes the evolution of state variables, driving input response and coupling between states.

[0085] When performing a delay embedding transformation on the operating condition feature vector, two key parameters need to be determined first: the time delay parameter and the embedding dimension parameter. The time delay parameter is determined using the mutual information function method. The mutual information values ​​of the operating condition feature vector sequence are calculated at different delay step sizes, and the delay step size corresponding to the first minimum value of the mutual information function is selected as the optimal time delay. Mutual information functions can effectively capture the nonlinear correlations between sequences and have stronger adaptability to nonlinear systems compared to traditional autocorrelation function methods. The embedding dimension parameter is determined using the correlation dimension method. By calculating the correlation integral between trajectory points in phase space, a saturation curve of the correlation dimension as a function of the embedding dimension is plotted. The embedding dimension corresponding to the curve approaching saturation is taken as the optimal embedding dimension. Using the two parameters mentioned above, the one-dimensional working condition feature vector time series is reconstructed into a set of trajectory points in a high-dimensional phase space according to the Takens embedding theorem. Each trajectory point consists of the current time and its preceding points. The eigenvector values ​​at each time delay are used to map the dynamic information of the original one-dimensional sequence into a high-dimensional phase space, thus providing a foundation for subsequent topological analysis.

[0086] Performing topological invariant analysis on the obtained trajectory point set is a core step in constructing the structural framework of the nonlinear dynamic model. When extracting the Lyapunov exponent spectrum of the trajectory point set, the Gram-Schmidt orthogonalization method is used to iteratively evolve and normalize the tangent vectors of adjacent trajectory points, calculating the average logarithmic stretching rate in each direction to obtain the complete Lyapunov exponent spectrum. ,in For the first The Lyapunov indices are sorted in descending order. The largest Lyapunov index... When the system exhibits chaotic characteristics, it indicates that the fracturing load dynamics system is sensitive to initial conditions, necessitating the introduction of nonlinear coupling terms into the model. When all exponents are negative, the system is in a stable attractor state, allowing for a relatively simplified state-space structure. The stability domain boundary is determined by calculating the critical parameter range where the Lyapunov exponents transition from negative to positive, providing a basis for safety constraints in subsequent energy scheduling.

[0087] The extraction of attractor geometry is achieved by calculating the attractor's fractal dimension and phase space volume evolution characteristics. Fractal dimension Box counting is used to estimate the effective degrees of freedom occupied by the attractor in phase space, thereby determining the minimum necessary dimension of the state variables. The coupling form is determined by analyzing the projection shape of the phase space trajectory on different coordinate planes to identify whether nonlinear coupling, time-delay coupling, or parameter modulation coupling exists between the state components. Based on the above analysis results, a structural framework for the state-space expression is generated, clarifying the state variables. The dimensional composition and physical correspondence of each component, as well as the coupling function. The nonlinear structural form.

[0088] In the parameterization stage, the state equations and observation equations in the structural framework are explicitly represented. The state equations describe the state variables. The time evolution pattern is in the form of ,in Let be the vector of undetermined parameters in the state equation. To drive the input, This is the index for the discrete time step. The observation equation describes the mapping between state variables and measurable power demand, in the form of: ,in For the first Power demand forecast at time t. For the observation mapping function, This represents the vector of undetermined parameters in the observation equations. The fracturing operation parameters serve as the driving input terms. By introducing state equations, including controllable construction parameters such as pumping displacement sequence and sand ratio sequence, the model can respond to the driving effect of construction operations on load dynamics and reflect the influence mechanism of human control factors on the evolution of power demand during fracturing.

[0089] When using a recursive identification strategy to estimate the parameters to be determined online, the weighted sum of squared prediction errors is used as the cost function. Defined as ,in For the first The measured power value at time [time]. This represents the total number of observation points within the current identification window. The time-varying weighting factor (forgetting factor) satisfies Introducing a time-varying weighting factor The purpose is to apply exponential decay processing to historical observation data, so that the contribution of historical data further back in time to parameter estimation is smaller, thereby giving the model the adaptive ability to track changes in operating conditions. When fracturing operations enter different stages of operating conditions (such as stepwise changes in displacement, rapid increases in sand ratio, etc.), The value can be dynamically adjusted according to the rate of change of operating conditions; it should be appropriately reduced when the operating conditions change drastically. The value is increased to accelerate parameter convergence; it should be appropriately increased when the operating conditions are stable. Values ​​are used to enhance the stability of parameter estimation.

[0090] The parameter iterative update adopts a recursive least squares framework. When new observation data arrives, it is not necessary to reprocess all historical data; instead, the parameter estimates and covariance matrix are corrected only through an incremental update formula, ensuring the real-time nature of online identification. For The nonlinear coupling terms in the model are approximated using a local linearization method. A first-order Taylor expansion of the nonlinear function is performed in the neighborhood of the current operating point, transforming the nonlinear parameter identification problem into a local linear regression problem. This reduces computational complexity while preserving the effective descriptive ability of the nonlinear dynamics. After identification, a complete nonlinear dynamic model is obtained, encompassing the evolution of state variables, the characteristics of the driving input response, and the coupling effects between states. This model can be directly used for rolling prediction of power demand evolution trajectories across multiple time windows, providing a reliable predictive basis for generating the globally optimal energy dispatch trajectory.

[0091] In actual fracturing operations, due to the complex and variable formation conditions, the statistical characteristics of the operating condition feature vector drift with changes in drilling depth, formation lithology, and fracturing fluid formulation. Therefore, the root mean square value of the prediction error is continuously monitored during model operation. When the error exceeds the preset tolerance zone, a parameter re-identification process is automatically triggered to ensure that the nonlinear dynamic model always maintains a high degree of consistency with the current fracturing conditions, providing accurate power demand prediction information for upper-level energy dispatch decisions.

[0092] In one optional implementation, by solving a multi-objective optimization problem that includes energy constraints, power constraints, and lifetime constraints, a globally optimal energy scheduling trajectory covering pre-charge timing, discharge depth allocation, and main power supply coordination ratio is generated, including:

[0093] Based on the power demand evolution trajectory, the power fluctuation amplitude sequence and duration sequence of fracturing operations are extracted, the releaseable energy of the supercapacitor bank at different discharge depths and the corresponding cycle life loss rate are calculated, and a nonlinear mapping relationship between discharge depth and cycle life loss rate is constructed.

[0094] The nonlinear mapping relationship is used as a mathematical representation of the lifetime constraint. An energy constraint inequality is established by combining the upper and lower limits of the state of charge of the supercapacitor bank. A power constraint inequality is established by combining the physical limits of charging and discharging power. The peak load suppression target of the main power supply and the energy cycle efficiency target of the supercapacitor bank are introduced to construct a multi-objective optimization problem.

[0095] The multi-objective optimization problem is solved in layers. In the outer layer optimization, the trade-off coefficient between energy utilization and lifetime loss is determined by Pareto front search. In the inner layer optimization, the optimal decision sequence that satisfies all constraints is solved based on the trade-off coefficient.

[0096] In the inner optimization, the decision variables are discretized into a set of pre-charge start-up times, a discharge depth allocation vector for each time period, and a main power supply coordination ratio vector. A dynamic programming strategy is used to solve the problem recursively for each time period. The final value of the state of charge of the previous time period is used as the initial value of the state transition of the current time period. The impact of the decision of the current time period on the feasible region of the subsequent time periods is calculated through the state transition equation. Finally, the global optimal energy scheduling trajectory is obtained by backtracking.

[0097] like Figure 3 As shown, the method includes:

[0098] This figure presents the optimal scheduling results that satisfy energy, power, and lifetime constraints. The horizontal axis represents 12 scheduling periods, and the vertical axis represents power (MW). The stacked bars represent the output distribution between the supercapacitor (green) and the main power source (blue) in each period, and the broken line represents the total load demand. In periods 2-12 (peak power demand period), the supercapacitor discharge depth reaches 2.8 MW, and the main power source output drops to 0.2 MW, effectively suppressing the peak fluctuation of the main power source. This trajectory is solved recursively through dynamic programming, using the final value of the state of charge in the previous period as the initial value for the next period, and backtracking to obtain the globally optimal decision sequence. Simultaneously, the figure demonstrates a rolling correction mechanism—when the actual load deviates from the forecast, the scheduling ratio in subsequent periods can be adjusted online to ensure that the constraints are always met. This trajectory balances the smoothness of the main power source (minimizing variance) and the lifetime loss of the supercapacitor (cumulative loss ≤ health threshold), providing a quantitative basis for adaptive power allocation.

[0099] After obtaining the power demand evolution trajectory for multiple future time windows, the power fluctuation amplitude sequence and duration sequence of fracturing operations are extracted. Specifically, the power demand evolution trajectory is segmented and statistically analyzed by time period. The peak-to-valley difference of each power fluctuation is recorded as an element of the amplitude sequence, and the continuous duration of each power level remaining in the high-power range is recorded as an element of the duration sequence. These two sets of sequences together characterize the energy demand distribution characteristics of fracturing operations over time, providing input for the subsequent formulation of supercapacitor discharge strategies.

[0100] A significant nonlinear relationship exists between the releaseable energy of a supercapacitor bank at different depths of discharge and its cycle life loss rate. Let the depth of discharge of the supercapacitor bank be... (Values ​​range from 0 to 1, representing the discharge ratio from the upper limit of the state of charge to the lower limit of the state of charge), at the depth of discharge. Below, the energy that a supercapacitor bank can release in a single cycle With rated total energy storage Between ,in This refers to the round-trip charging and discharging efficiency of the supercapacitor bank. Simultaneously, it refers to the cycle life loss rate. The discharge depth increases at an accelerating rate. By performing nonlinear fitting on historical cycle life test data of the supercapacitor bank, a [system / mechanism] was established. about The power function mapping relationship, i.e. ,in and A time-varying parameter that takes into account the cumulative effects of aging. The lifetime decay coefficient. and life decay index The following experimental conditions were used for calibration: Accelerated aging tests were conducted on the supercapacitor bank at an ambient temperature of 25±2℃ and a charge / discharge rate of 2C. The relationship between the number of cycles and capacity decay at different depths of discharge (0.2, 0.4, 0.6, 0.8, 1.0) was recorded, and the results were obtained through nonlinear least squares fitting. The typical value range is 0.0008. The typical value range is 2.3. This varies depending on the number of cycles accumulated in the supercapacitor bank. As the time-varying parameters increase, they are dynamically updated according to the following rules: , ,in Rated cycle life. Updated in real time via online identification module. This allows the calculation of the life loss rate to reflect the objective law that the relative life loss of capacitors increases with the degree of aging, even at the same depth of discharge. This nonlinear mapping relationship serves as a mathematical representation of the life constraint, ensuring that the quantitative assessment of the supercapacitor bank's life consumption during the optimization process has a physical basis.

[0101] When establishing the constraint system for the multi-objective optimization problem, the upper and lower limits of the state of charge (SOC) of the supercapacitor bank are considered. and Establish an energy-constrained inequality, requiring that at any time... State of charge satisfy This prevents overcharging or over-discharging from damaging the capacitor itself. It also considers the physical limits of the charging and discharging power of the supercapacitor bank. Establish a power constraint inequality to require that the actual charging and discharging power of the supercapacitor bank at any given time be... The absolute value does not exceed At the objective function level, two optimization objectives are introduced: one is the peak load suppression objective of the main power supply, that is, minimizing the peak power of the main power supply throughout the entire scheduling cycle. The first objective is to smooth out peaks and troughs in the main power supply curve by timely discharging the supercapacitor bank during power surges. The second objective is to maximize the overall energy utilization efficiency of the supercapacitor bank during the scheduling cycle and reduce energy losses caused by ineffective charge-discharge cycles. These two objectives are inherently competitive and require a multi-objective optimization framework for balance.

[0102] A hierarchical solution strategy is adopted for the multi-objective optimization problem. The outer-layer optimization determines the trade-off coefficient between the main power supply peak load suppression objective and the supercapacitor bank lifetime loss through Pareto front search. The two objectives are weighted and combined into a single scalar objective function. ,in This is the normalized target value for the peak load of the main power supply. This represents the normalized lifetime loss target. Outer layer optimization is achieved through... Within the range of values, uniform sampling or adaptive search is performed to enumerate a set of Pareto optimal solutions. Then, considering the priority requirements of equipment life protection and power supply stability in actual fracturing operations, the final trade-off coefficient is selected from the Pareto front. .

[0103] Inner layer optimization is based on the tradeoffs determined by the outer layer. Solve for the optimal decision sequence that satisfies all constraints. Divide the scheduling period into... The decision variables are discretized into three categories over a given time period: the set of pre-charge start-up times. (Including the trigger time of each precharge) and the depth of discharge allocation vector for each time period. (Each element corresponds to the depth of discharge for the corresponding time period) and the main power supply coordination ratio vector (Each element represents the proportion of power supplied by the main power source during the corresponding time period). These three types of decision variables jointly determine the energy allocation scheme between the supercapacitor bank and the main power source during each time period.

[0104] A dynamic programming strategy is employed to recursively solve the inner-level optimization problem on a time-by-time basis. This is based on the state of charge of the supercapacitor bank. As a state variable, the discharge depth at each time period ( (For time period index) and mains power coordination ratio As decision variables, construct the state transition equation. In the... Time period, given initial state of charge Power requirements Depth of discharge Coordination ratio with main power supply The final value of the charged state at the end of this period is calculated using the state transition equation. ,Right now ,in For the first Duration of the time period. As the first Initial value of state transition for each time period The decision is then passed on to the next time period for recursive calculation. During the recursive process, the cumulative cost function value is calculated for each feasible decision combination in each time period, and the optimal decision path is recorded. At the same time, the impact of the current time period's decision on the feasible domain of the charged state in subsequent time periods is predicted through the state transition equation. When a decision causes the charged state in subsequent time periods to exceed the boundary, the decision is marked as infeasible and removed from the candidate set, thereby ensuring the satisfaction of global constraints.

[0105] After completing the forward recursion for all time periods, starting from the optimal cumulative cost of the final time period, backtracking along the recorded optimal decision path, and sequentially extracting the optimal discharge depth for each time period. Main power supply coordination ratio The pre-charging start time, combined with the pre-charging start time, ultimately yields a globally optimal energy scheduling trajectory covering the entire scheduling cycle. This trajectory clearly defines the charging and discharging timing and magnitude of the supercapacitor bank in each time period, as well as the output ratio of the main power supply, providing a complete execution basis for the precise control of the subsequent adaptive power allocation unit. The pre-charging timing is determined based on the time window preceding high-power surge events identified in the power demand evolution trajectory. By charging the supercapacitor bank to the target state of charge in advance, sufficient discharge energy reserves are ensured when the surge event occurs, while avoiding excessive energy loss due to self-discharge during the waiting period caused by premature pre-charging.

[0106] After the global optimal energy dispatch trajectory is generated, a rolling correction mechanism is needed in the actual execution process. Based on the real-time collected power demand deviation and state of charge deviation, the decision sequence for subsequent periods is locally re-optimized to cope with the random disturbances and prediction error accumulation of fracturing conditions, and to ensure the dynamic adaptability and robustness of the dispatch trajectory.

[0107] In one optional implementation, the nonlinear mapping relationship is used as a mathematical representation of the lifetime constraint. An energy constraint inequality is established by combining the upper and lower limits of the supercapacitor's state of charge, and a power constraint inequality is established by combining the physical limits of charge and discharge power. Furthermore, the peak load suppression objective of the main power supply and the energy cycle efficiency objective of the supercapacitor are introduced to construct a multi-objective optimization problem, including:

[0108] The discharge depth and cycle lifetime loss rate in the nonlinear mapping relationship are transformed into a cumulative lifetime loss expression. The lifetime loss at each moment in the entire scheduling cycle is accumulated, and a lifetime constraint inequality is established to limit the cumulative lifetime loss to not exceed the health threshold.

[0109] Based on the cumulative lifetime loss expression, the contribution weight of the discharge depth to the total lifetime loss at each moment is calculated, the lifetime loss sensitive period is identified, the upper and lower limits of the supercapacitor bank's state of charge are converted into the energy reserve requirements and energy retention requirements of the lifetime loss sensitive period, and an energy constraint inequality is established.

[0110] Extract the dynamic boundary of the physical limit of charging and discharging power with respect to the state of charge, and establish a power constraint inequality to constrain the charging and discharging power at each moment to be within the feasible region determined by the dynamic boundary.

[0111] The peak load suppression objective of the main power supply is defined as the expression for minimizing the variance of the power supply power fluctuation, and the energy cycle efficiency objective of the supercapacitor bank is defined as the expression for minimizing the charging and discharging energy loss. The peak load suppression objective of the main power supply and the energy cycle efficiency objective of the supercapacitor bank are assigned weight coefficients and weighted and summed to form a comprehensive objective. The comprehensive objective is combined with the lifetime constraint inequality, the energy constraint inequality and the power constraint inequality to construct a multi-objective optimization problem.

[0112] like Figure 4 As shown, the method includes:

[0113] After constructing the nonlinear dynamic model and predicting the power demand evolution trajectory, it is necessary to integrate the physical characteristic constraints of the supercapacitor bank and the multi-objective optimization objective into a solvable mathematical programming framework. The starting point for this process is to transform the nonlinear mapping relationship between the depth of discharge and cycle life loss into a cumulative lifetime loss expression that can be directly applied in the optimization problem.

[0114] Within the scheduling cycle, for each time period Its corresponding cycle life loss rate From the lifetime decay coefficient Lifespan Decline Index Determined according to the power law relationship, that is ,in For the first The depth of discharge during different time periods. The cycle life attenuation rate at different time periods is not only related to the depth of discharge. It is also related to the accumulated number of cycles at the start time of that period. This ensures that the more severe the aging, the higher the relative lifespan loss for the same depth of discharge, consistent with the actual aging characteristics of supercapacitors. The cumulative lifespan loss is obtained by summing the lifespan losses for all periods throughout the entire scheduling cycle. , expressed as To ensure that the supercapacitor bank does not age excessively throughout its service life, a health threshold is introduced. Establish life-constrained inequalities This threshold is determined based on a combination of the factory-rated number of cycles for the supercapacitor bank and the current accumulated usage losses, and is updated by the online identification module before the start of each scheduling cycle.

[0115] After determining the cumulative lifetime loss expression, the contribution weight of discharge depth to total lifetime loss in each time period is further analyzed to identify lifetime loss-sensitive periods. For the ... Contribution weight of each time period Defined as the ratio of lifetime loss during that period to cumulative lifetime loss, i.e. .when Exceeding the preset sensitivity threshold At that time, the first The time period is marked as a lifespan-sensitive period, and the sensitive period index set is denoted as . .for The power impact amplitude and discharge depth of fracturing operations are relatively large. Therefore, it is necessary to ensure that the supercapacitor bank has sufficient energy reserves before such periods and to retain necessary energy margins after such periods to cope with subsequent continuous impacts.

[0116] Based on the above analysis, the upper and lower limits of the state of charge are... and This is transformed into an energy-constrained inequality for periods sensitive to lifetime depletion. For each sensitive period... At the start of the time period, the state of charge must not be lower than the lower limit of energy reserves. This value is derived from the predicted power demand. Duration of the period Combined with the rated total energy storage of the supercapacitor bank and charge / discharge round-trip efficiency The calculation yields the following result, which is expressed as: At the end of the time period, the state of charge must not be lower than the lower limit of energy retention. This is to prevent the supercapacitor bank from entering an over-discharge state. During non-sensitive periods, the state of charge is still subject to general constraints. and The limitations ensure that the state of charge remains within the safe operating range. The initial and final states of charge in each time period are interconnected through energy conservation relationships, forming a set of energy constraint inequalities covering the entire scheduling cycle.

[0117] Establishing the power constraint inequality requires considering the physical limits of the charging and discharging power of the supercapacitor bank. The dynamic characteristics of the state of charge. Approaching the upper limit of the state of charge. At this point, the charging power is significantly reduced due to the limitation of terminal voltage saturation; when the state of charge is close to the lower limit... At the same time, the discharge power is also constrained by the voltage drop at the terminals. A dynamic power boundary function is introduced. The upper limit of usable power, which varies with the state of charge, is specifically obtained by fitting the parameters (internal resistance, capacitance, and terminal voltage characteristic curves) of the equivalent circuit model. The power constraint inequality is expressed as follows for any time... All meet ,in for The actual charging and discharging power of the supercapacitor bank at any given time is represented by a negative value indicating the charging direction and a positive value indicating the discharging direction. This constraint limits the power decision variables at each time point to a feasible region determined by the current state of charge, thus preventing damage to the supercapacitor bank or control failure due to power overload.

[0118] After the constraint system is established, two objective functions are defined for the multi-objective optimization problem. The objective of suppressing peak load of the main power supply is mathematically represented by minimizing the variance of the main power supply power fluctuation. Let the main power supply be defined in the first... The output power during the time period is The average power of the main power supply during the scheduling period is The target for suppressing peak load of the main power supply is... Defined as By minimizing this variance, the output power of the main power supply is smoothed, reducing the mechanical stress and fuel consumption caused by frequent and large fluctuations in the main power supply. The energy cycle efficiency target of the supercapacitor bank is characterized by minimizing the charge and discharge energy loss. The round-trip charge and discharge loss comes from internal resistance heat dissipation. Energy loss during a period of time Compared with the absolute value of charging and discharging energy and round-trip efficiency during this period Related, definition ,in For the first Average charge / discharge power of the supercapacitor bank over a given period. This is achieved by minimizing... This can reduce unnecessary charge and discharge cycles and extend the actual service life of the supercapacitor bank while meeting power requirements.

[0119] After normalizing the two objective functions, a tradeoff coefficient is introduced. A comprehensive objective function is constructed by weighting and summing the peak load suppression target of the main power supply and the energy cycle efficiency target. ,in and All have completed dimensionless normalization through their respective historical maximum values. The range of values ​​is When fracturing operations require high stability of the main power supply, Choose the larger value to prioritize ensuring smooth mains power output; when the operating cycle is long and supercapacitor bank lifespan management is of higher priority. Choose a smaller value to focus on reducing charging and discharging losses. The final selected value The decision was made by combining Pareto front analysis with a trade-off of on-site operational priorities.

[0120] The comprehensive objective function Inequality with lifetime constraints The energy constraint inequalities and the power constraint inequalities are combined to form a complete multi-objective optimization problem. The decision variable is the discharge depth allocation vector for each time period. Main power supply cooperative proportional vector and pre-charge start-up time set This optimization problem belongs to mixed integer nonlinear programming with nonlinear constraints. It can be solved offline before the start of each scheduling cycle using sequential quadratic programming or particle swarm optimization to obtain the globally optimal energy scheduling trajectory. During execution, the scheduling trajectory is corrected online through a rolling time-domain control framework to address the risk of constraint violation caused by deviations between actual fracturing conditions and predictions.

[0121] In one optional implementation, the equivalent circuit model parameters of the supercapacitor bank are updated in real time using online identification technology. When the drift of the model parameters is detected to exceed an adaptive threshold, a joint decision-making scheme is generated, including fault warning, topology reconfiguration instructions, and scheduling strategy compensation.

[0122] The terminal voltage and current responses of the supercapacitor bank are collected, and the equivalent series resistance, equivalent parallel capacitance and leakage resistance parameters in the equivalent circuit model are calculated through online identification technology. The calculated parameter values ​​are used as the parameters of the equivalent circuit model updated in real time.

[0123] Calculate the parameter drift of the real-time updated equivalent circuit model parameters relative to the initial calibration value, extract the drift rate and drift acceleration based on the time series features of the parameter drift, and dynamically generate an adaptive threshold according to the coupling relationship between the drift rate and the drift acceleration;

[0124] When the parameter drift exceeds the adaptive threshold, the contribution ratio of each model parameter in the parameter drift is analyzed, the dominant degradation parameter is identified, and fault warning information is generated based on the physical meaning of the dominant degradation parameter.

[0125] Based on the fault warning information, the location of the degraded individual cell in the supercapacitor bank is located, a topology reconfiguration instruction is generated to isolate the individual cell location and reconfigure the electrical connection path, the power output capacity attenuation coefficient and the energy reserve attenuation coefficient are calculated based on the equivalent parameters of the reconfigured topology, and a scheduling strategy compensation scheme is generated based on the power output capacity attenuation coefficient and the energy reserve attenuation coefficient.

[0126] The fault warning information, the topology reconfiguration instruction, and the scheduling strategy compensation scheme are combined to form a joint decision-making scheme.

[0127] The terminal voltage and current response signals of the supercapacitor bank are collected during operation. The sampling frequency must meet the requirements for capturing the transient processes of capacitor charging and discharging, typically not less than 1kHz. The online identification technology employs recursive least squares or an extended Kalman filter framework to represent the equivalent circuit model of the supercapacitor bank as including equivalent series resistance. Equivalent parallel capacitance and leakage resistance The three-parameter model is used. Within each identification cycle, an observation equation is established based on the time-domain relationship between the terminal voltage response and the current response. This equation is then iteratively solved by minimizing the weighted cumulative cost of the voltage prediction error. The current optimal estimate is obtained, and the solution result is stored in the parameter cache queue as the parameters of the equivalent circuit model updated in real time, for subsequent drift analysis.

[0128] For the parameters of the equivalent circuit model updated in real time, the absolute and relative deviations of each parameter relative to the initial factory calibration values ​​are calculated to obtain the parameter drift sequence. Let... The initial calibration value is The current identification value is ,but The drift amount is Similarly, The drift amount is , The drift amount is The drift rate was extracted by performing first-order and second-order differencing on the time series of each drift. With drift acceleration The drift rate reflects the current rate of parameter degradation, while the drift acceleration characterizes whether the degradation trend is intensifying. Adaptive threshold. in accordance with and The coupling relationship is dynamically generated; specifically, when A larger threshold indicates accelerated degradation, and the threshold should be tightened to trigger an early warning; when... When the threshold value is close to zero or negative, the degradation tends to plateau, and the threshold can be appropriately relaxed to avoid false alarms. The rule for generating the adaptive threshold can be expressed as follows: ,in As the baseline threshold, and After taking the absolute value, it is used in the calculation. Limiting calculation ensures The value should always be positive and not lower than 20% of the baseline threshold to avoid false alarms caused by an excessively small threshold or a negative threshold.

[0129] When the drift of any parameter exceeds the corresponding adaptive threshold Then, the fault analysis process begins. This involves analyzing the contribution percentage of each model parameter drift to the overall degradation index, and then analyzing the overall degradation index. Defined as the weighted sum of the normalized drift values ​​of each parameter, i.e. ,in These are the degradation weighting coefficients for equivalent series resistance, equivalent parallel capacitance, and leakage resistance, respectively. The sum of these three is 1, and the weights are pre-calibrated based on prior knowledge of the degree of influence of each parameter on power supply performance. The parameter with the largest contribution is identified as the dominant degradation parameter. When degradation is dominant, the physical meaning corresponds to aging of the electrode material or increased contact resistance, manifested as increased internal resistance and decreased charge / discharge efficiency. The fault warning information is marked as "Internal Resistance Exceeds Limit Warning" and includes the current status. The numerical value and its percentage deviation from the initial value; When degradation is dominant, the physical meaning corresponds to electrolyte loss or deterioration of the electrode double layer structure, which manifests as a decrease in energy storage capacity. The fault warning information is marked as "capacity decay warning". During dominant degradation, the physical meaning corresponds to increased self-discharge due to insulation deterioration, and the fault warning information is marked as "Self-discharge exceeding limit warning". The warning information also records the timestamp of the degradation occurrence, the current operating condition number, and the estimated remaining usable life, so that maintenance personnel can quickly locate the problem.

[0130] Based on the dominant degradation parameter types and their quantification levels identified in the fault warning information, and combined with the individual cell consistency monitoring data of the supercapacitor bank, the location of the degraded cells is determined. Individual cell consistency monitoring is achieved by comparing the mean deviation of the terminal voltage and the internal resistance dispersion coefficient of each cell; cells with deviations exceeding twice the standard deviation of the group mean are marked as degraded cells. For degraded cells, a topology reconfiguration instruction is generated, containing the switching action sequence for isolating the target cell and a topology scheme for reconfiguring electrical connection paths. Topology reconfiguration can choose between two modes: bypass isolation or series-parallel reconfiguration. Bypass isolation is suitable for cases with a small number of cells and minor degradation, where the degraded cell is short-circuited by closing a bypass switch. Series-parallel reconfiguration is suitable for cases where multiple cells degrade simultaneously, where the group-level voltage and capacity balance is maintained by adjusting the series-parallel combination of the remaining healthy cells. After reconfiguration, the power output capacity attenuation coefficient of the supercapacitor bank is recalculated based on the equivalent parameters of the reconfigured topology. With energy storage decay coefficient ,in , , and These represent the equivalent maximum power and equivalent total energy storage of the reconfigured topology, respectively. and This is the rated design value.

[0131] according to and The core of the generated scheduling strategy compensation scheme is to multiply the power commands involving supercapacitors in the globally optimal energy scheduling trajectory by a certain factor in subsequent scheduling cycles. A derating process is implemented, while the lower limit of the state of charge constraint is appropriately raised to reserve additional energy margin to compensate for the reduction in available energy caused by capacity decay. Specifically, if Below the preset capacity compensation trigger threshold If so, positive compensation of the main power supply coordination ratio will be initiated, increasing the output ratio of the main power supply during high-power periods based on the original scheduling trajectory, in order to compensate for the power supply gap caused by insufficient energy reserves of the supercapacitor bank. If Below the power compensation trigger threshold This applies a hard limit to the peak power command of the supercapacitor bank to prevent over-discharge from accelerating the degradation of remaining cells. The compensation scheme also includes a re-optimization trigger flag, which is triggered when... or When the attenuation exceeds 5%, the global scheduling trajectory is re-solved. Based on the reconstructed equivalent parameters, the optimal scheduling trajectory that satisfies the energy constraint, power constraint and lifetime constraint is regenerated to ensure that the power supply scheme still has global optimality after topology change.

[0132] The fault warning information, topology reconfiguration instructions, and scheduling strategy compensation scheme are integrated into a joint decision-making scheme. The three parts are strictly aligned in timestamps to ensure consistency in execution order. The execution results of the joint decision-making scheme, including the actual power response after topology reconfiguration, state-of-charge trajectory, and power supply error, are fed back to the nonlinear dynamic model in real time. This updates the initial conditions of the model's state variables and the driving input boundaries, forming a complete closed-loop optimization link from parameter identification to model update to scheduling optimization. This enables the entire power supply control system to maintain stable and efficient fracturing power supply capability even when the supercapacitor bank experiences performance degradation.

[0133] In one optional implementation, the equivalent series resistance, equivalent parallel capacitance, and leakage resistance parameters in the equivalent circuit model are calculated using online identification technology, and the calculated parameter values ​​are used as the parameters of the equivalent circuit model updated in real time, including:

[0134] The amplitude of the terminal voltage jump and the current step response of the supercapacitor bank during the charging and discharging transition are collected, and the current resistance value of the equivalent series resistance is calculated based on the ratio of the amplitude of the terminal voltage jump to the current step response.

[0135] The terminal voltage rise curve of the supercapacitor bank during the constant current charging stage is collected. The voltage change rate sequence is obtained by performing time-domain differentiation on the terminal voltage rise curve. The real-time capacitance value of the equivalent parallel capacitor is inverted based on the numerical relationship between the voltage change rate sequence and the charging current.

[0136] The self-discharge decay curve of the terminal voltage of the supercapacitor bank during the resting stage is collected. The decay time constant is extracted by exponential fitting of the terminal voltage self-discharge decay curve. The current value of the leakage resistance is calculated based on the decay time constant and the real-time capacitance value.

[0137] The current resistance value of the equivalent series resistor, the real-time capacitance value of the equivalent parallel capacitor, and the current resistance value of the leakage resistor are timestamped to establish a parameter update record sequence. The parameter values ​​of multiple consecutive sampling periods in the parameter update record sequence are smoothed by a sliding window to eliminate the interference of transient noise on parameter identification. The filtered parameter values ​​are used as the parameters of the equivalent circuit model after real-time update.

[0138] In the online parameter identification process of supercapacitor banks, the real-time calculation of the equivalent series resistance relies on the physical relationship between the voltage jump and the current step response during the transient transition between charging and discharging. When the supercapacitor bank switches from charging to discharging or vice versa, the voltage jump is proportional to the current step amplitude at the moment of switching due to the presence of the equivalent series resistance. Specifically, the steady-state values ​​of the voltage jump are collected before and after the switching moment, and the difference between the two values ​​is calculated to obtain the voltage jump amplitude. Simultaneously record the current step at the switching moment. The current resistance value of the equivalent series resistance is... satisfy To ensure acquisition accuracy, the sampling time window for the voltage jump amplitude needs to be controlled on the order of microseconds to avoid contamination of the jump signal by the charging and discharging process of the capacitor. In actual engineering deployments, the fracturing operation environment has strong electromagnetic interference. Therefore, when acquiring the voltage jump signal, a high-speed differential sampling circuit is required to suppress the influence of common-mode noise and ensure accuracy. The measurement error is within a reasonable range.

[0139] Real-time capacitance identification of the equivalent parallel capacitor is based on the terminal voltage ramp-up characteristics during constant current charging. Under constant current charging conditions, the current flowing into the supercapacitor and the charging current are compared. Approximately equal (ignoring the current shunting in the leakage branch), the terminal voltage increases linearly with time, and its slope is inversely proportional to the capacitance. The collected terminal voltage rise curves are analyzed. Perform time-domain differentiation to obtain the voltage change rate sequence. Under ideal constant current conditions, It should be a constant, but in practice, it will change slowly due to temperature drift and nonlinear effects. Therefore, it is taken within the sliding window. mean As a stable estimator, the real-time capacitance value of the equivalent parallel capacitor. Inverse equation: Differentiation operations are highly sensitive to sampling noise. Before differentiating the voltage rise curve, a low-pass filter must be applied to the original voltage sequence. The selection of the cutoff frequency should balance the signal dynamic response bandwidth and noise suppression effect. Charging current in fracturing operations... The stability of the capacitance value directly affects the accuracy of capacitance identification, therefore it is necessary to use a current sensor to... Perform high-precision synchronous data acquisition and maintain strict time synchronization with voltage acquisition.

[0140] The current resistance value of the leakage resistor is identified using the self-discharge decay characteristic of the supercapacitor bank's terminal voltage during the resting phase. The resting phase refers to the period during which the supercapacitor bank is disconnected from the external circuit and self-discharges solely through its internal leakage resistor. During this phase, the terminal voltage... The decay law approximately satisfies the first-order exponential decay form, that is... ,in This represents the initial terminal voltage at the start of the resting period. The decay time constant is calculated by performing an exponential fit on the acquired terminal voltage self-discharge decay curve. Combined with the real-time capacitance value of the identified equivalent parallel capacitor. Current leakage resistance value Calculated by the following formula: In the exponential fitting process, the selection of the fitting interval is crucial: During the initial resting period, the terminal voltage decays rapidly, and due to the residual effect of the equivalent series resistance, it is recommended to skip the data from the first few seconds after resting and start fitting from the linear logarithmic segment where the terminal voltage tends to stabilize, in order to improve accuracy. The estimation accuracy is improved. During the intervals between fracturing operations, the static data acquisition process can be automatically triggered during equipment standby periods to ensure the periodic updating of leakage resistance identification data.

[0141] After completing the real-time calculation of the three parameters, the current resistance value of the equivalent series resistance is determined. Real-time capacitance value of equivalent parallel capacitor and the current value of leakage resistance Add timestamps respectively This creates a time-indexed sequence of parameter update records. The purpose of the timestamps is to provide a time reference for subsequent calculations of parameter drift rates and accelerations, and also to facilitate tracing historical data and pinpointing the time of parameter mutations when parameters become abnormal. The parameter update record sequence is stored in a circular buffer, with a buffer length of... The value is set based on a combination of the identification period and the system response time requirements, with a typical range of tens to hundreds of sampling periods.

[0142] Sliding window smoothing filtering is a key step in eliminating transient noise interference with parameter identification, and it is applied to continuous parameters in the update record sequence. The parameter values ​​for each sampling period are used to independently apply sliding window averaging filtering to the three parameters. Taking the equivalent series resistance as an example, the parameter values ​​within the window are... The resistance at each sampling point is Filtered output value satisfy Window length The selection of the window size requires a trade-off between noise suppression and real-time parameter tracking: an excessively long window can lead to over-smoothing of the actual parameter drift, delaying fault detection response; a window that is too short cannot effectively filter out transient noise. During the high-power impact phase of fracturing operations, it is recommended to appropriately shorten the window size. To improve parameter tracking sensitivity; during the steady-state power supply phase, the time can be appropriately extended. To obtain more stable parameter estimates.

[0143] The sliding window smoothing filter for the equivalent parallel capacitance and leakage resistance uses the same processing framework as the equivalent series resistance, yielding the filtered estimated values ​​of the equivalent parallel capacitance. and the filtered leakage resistance estimate The three filtered parameter values ​​together constitute the parameter set of the equivalent circuit model updated in real time, serving as the input for subsequent parameter drift detection, comprehensive degradation index calculation, and topology reconstruction decisions. At the end of each identification cycle, , and The parameters are written to the equivalent circuit model parameter register for real-time reading by the energy scheduling module and the fault diagnosis module. Updates to the parameter register employ atomic write operations to avoid read-write contention during parameter writing, ensuring that scheduling calculations are always based on a consistent parameter state.

[0144] At the engineering implementation level, the identification timing of the above three types of parameters is closely coupled with the fracturing operation conditions. Equivalent series resistance identification relies on charge / discharge switching events, which occur frequently during power regulation in fracturing operations, ensuring a high identification frequency. Equivalent parallel capacitance identification relies on the constant current charging phase, requiring sufficient constant current charging time to be reserved in the scheduling strategy. Leakage resistance identification relies on the static phase, typically performed during operation breaks or equipment standby, resulting in a relatively low identification frequency. The identification results of the three types of parameters are aligned through a unified timestamp mechanism, maintaining consistency on the timeline in the parameter update record sequence even with different identification frequencies, providing a standardized input data format for sliding window smoothing filtering.

[0145] A second aspect of the present invention provides an intelligent control system for power supply to oil well fracturing based on supercapacitors, comprising:

[0146] The acquisition and prediction unit is used to acquire multi-source heterogeneous data from the fracturing operation site, perform multi-scale time-frequency decomposition on the multi-source heterogeneous data, extract the transient power impact characteristics and steady-state energy consumption characteristics of the fracturing operation, construct a nonlinear dynamic model of the fracturing load, and predict the power demand evolution trajectory of multiple future time windows based on the nonlinear dynamic model.

[0147] The energy scheduling unit is used to establish a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence based on the power demand evolution trajectory. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, it generates a globally optimal energy scheduling trajectory that covers pre-charging timing, discharge depth allocation and main power supply coordination ratio. During the execution process, the globally optimal energy scheduling trajectory is rolled over and corrected.

[0148] The power distribution unit is used to control the energy interaction between the supercapacitor bank and the main power supply through the adaptive power distribution unit according to the global optimal energy scheduling trajectory, so as to achieve precise power supply to the fracturing equipment.

[0149] The identification feedback unit is used to update the equivalent circuit model parameters of the supercapacitor bank in real time through online identification technology. When the drift of the model parameters is detected to exceed the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

[0150] A third aspect of the present invention provides an electronic device, comprising:

[0151] processor;

[0152] Memory used to store processor-executable instructions;

[0153] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0154] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0155] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart control method for power supply in oil well fracturing based on supercapacitors, characterized in that, include: Multi-source heterogeneous data from fracturing operations are collected, and multi-scale time-frequency decomposition is performed on the multi-source heterogeneous data to extract the transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of fracturing load is constructed, and the evolution trajectory of power demand in multiple future time windows is predicted based on the nonlinear dynamic model. Based on the power demand evolution trajectory, a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence is established. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, a globally optimal energy scheduling trajectory covering pre-charging timing, discharge depth allocation and main power supply coordination ratio is generated, and the globally optimal energy scheduling trajectory is rolled over during execution. Based on the global optimal energy scheduling trajectory, the energy interaction between the supercapacitor bank and the main power supply is controlled by the adaptive power distribution unit to achieve precise power supply to the fracturing equipment; The equivalent circuit model parameters of the supercapacitor bank are updated in real time by online identification technology. When the drift of the model parameters exceeds the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

2. The method according to claim 1, characterized in that, Multi-scale time-frequency decomposition is performed on the multi-source heterogeneous data to extract transient power impact characteristics and steady-state energy consumption characteristics of fracturing conditions. A nonlinear dynamic model of fracturing load is constructed, and based on the nonlinear dynamic model, the power demand evolution trajectory for multiple future time windows is predicted, including: Time alignment and sampling rate normalization are performed on multi-source heterogeneous data to construct a multi-dimensional data matrix under a unified time reference. Short-time Fourier transform and wavelet packet decomposition are performed on the multidimensional data matrix to obtain time-frequency representations at different scales. The amplitude envelope and duration of transient power impact events are identified from the time-frequency representations. The frequency domain dominant components of the steady-state energy consumption stage are extracted. The transient power impact features and steady-state energy consumption features are fused and mapped to generate the operating condition feature vector. Based on the aforementioned operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted by nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships. The current working condition feature vector is used as the initial state of the nonlinear dynamic model. The displacement sequence and sand ratio sequence in the fracturing construction plan are combined as future driving inputs. The state evolution equation of the nonlinear dynamic model in multiple future time windows is solved by numerical integration method to obtain the power demand evolution trajectory including power amplitude, rate of change and duration.

3. The method according to claim 2, characterized in that, Based on the aforementioned operating condition feature vector, a state-space reconstruction method is used to establish the phase space trajectory of the fracturing load. The evolution law of the phase space trajectory is fitted using nonlinear system identification technology to construct a nonlinear dynamic model that includes state variables, driving inputs, and coupling relationships. The working condition feature vector is subjected to a delay embedding transformation. The time delay parameter is determined by calculating the mutual information function, and the embedding dimension parameter is determined by calculating the correlation dimension. The one-dimensional working condition feature vector sequence is reconstructed into a set of trajectory points in a high-dimensional phase space using the time delay parameter and the embedding dimension parameter. A topological invariant analysis is performed on the trajectory point set to extract the Lyapunov exponent spectrum and attractor geometric features of the trajectory point set. The chaotic characteristics and stability domain boundary of the system are determined based on the Lyapunov exponent spectrum, and the state variable dimension and coupling form of the fracturing load dynamic system are determined based on the attractor geometric features, thereby generating a structural framework for the state space expression. The state equations and observation equations in the structural framework of the state-space expression are parameterized. The state equations describe the time evolution of state variables, and the observation equations describe the mapping relationship between state variables and measurable power demand. Fracturing construction parameters are introduced into the state equations as driving inputs. A recursive identification strategy is used to estimate the undetermined parameters in the state equation and the observation equation online. The parameter values ​​are updated iteratively by minimizing the cost function of the prediction error. A time-varying weighting factor is introduced to attenuate the historical observation data to track the changes in operating conditions, resulting in a nonlinear dynamic model that includes the evolution of state variables, driving input response and coupling between states.

4. The method according to claim 1, characterized in that, By solving a multi-objective optimization problem that includes energy constraints, power constraints, and lifetime constraints, a globally optimal energy scheduling trajectory is generated, encompassing pre-charge timing, discharge depth allocation, and main power supply coordination ratio. Based on the power demand evolution trajectory, the power fluctuation amplitude sequence and duration sequence of fracturing operations are extracted, the releaseable energy of the supercapacitor bank at different discharge depths and the corresponding cycle life loss rate are calculated, and a nonlinear mapping relationship between discharge depth and cycle life loss rate is constructed. The nonlinear mapping relationship is used as a mathematical representation of the lifetime constraint. An energy constraint inequality is established by combining the upper and lower limits of the state of charge of the supercapacitor bank. A power constraint inequality is established by combining the physical limits of charging and discharging power. The peak load suppression target of the main power supply and the energy cycle efficiency target of the supercapacitor bank are introduced to construct a multi-objective optimization problem. The multi-objective optimization problem is solved in layers. In the outer layer optimization, the trade-off coefficient between energy utilization and lifetime loss is determined by Pareto front search. In the inner layer optimization, the optimal decision sequence that satisfies all constraints is solved based on the trade-off coefficient. In the inner optimization, the decision variables are discretized into a set of pre-charge start-up times, a discharge depth allocation vector for each time period, and a main power supply coordination ratio vector. A dynamic programming strategy is used to solve the problem recursively for each time period. The final value of the state of charge of the previous time period is used as the initial value of the state transition of the current time period. The impact of the decision of the current time period on the feasible region of the subsequent time periods is calculated through the state transition equation. Finally, the global optimal energy scheduling trajectory is obtained by backtracking.

5. The method according to claim 4, characterized in that, Using the aforementioned nonlinear mapping relationship as a mathematical representation of lifetime constraints, energy constraint inequalities are established in conjunction with the upper and lower limits of the supercapacitor's state of charge, and power constraint inequalities are established in conjunction with the physical limits of charge and discharge power. Furthermore, by introducing the main power supply peak load suppression objective and the supercapacitor's energy cycle efficiency objective, a multi-objective optimization problem is constructed, including: The discharge depth and cycle lifetime loss rate in the nonlinear mapping relationship are transformed into a cumulative lifetime loss expression. The lifetime loss at each moment in the entire scheduling cycle is accumulated, and a lifetime constraint inequality is established to limit the cumulative lifetime loss to not exceed the health threshold. Based on the cumulative lifetime loss expression, the contribution weight of the discharge depth to the total lifetime loss at each moment is calculated, the lifetime loss sensitive period is identified, the upper and lower limits of the supercapacitor bank's state of charge are converted into the energy reserve requirements and energy retention requirements of the lifetime loss sensitive period, and an energy constraint inequality is established. Extract the dynamic boundary of the physical limit of charging and discharging power with respect to the state of charge, and establish a power constraint inequality to constrain the charging and discharging power at each moment to be within the feasible region determined by the dynamic boundary. The peak load suppression objective of the main power supply is defined as the expression for minimizing the power fluctuation variance of the main power supply, and the energy cycle efficiency objective of the supercapacitor bank is defined as the expression for minimizing the charging and discharging energy loss. The peak load suppression objective of the main power supply and the energy cycle efficiency objective of the supercapacitor bank are assigned weight coefficients and weighted summed to form a comprehensive objective. The comprehensive objective is combined with the lifetime constraint inequality, the energy constraint inequality and the power constraint inequality to construct a multi-objective optimization problem.

6. The method according to claim 1, characterized in that, The equivalent circuit model parameters of the supercapacitor bank are updated in real time using online identification technology. When the drift of the model parameters exceeds the adaptive threshold, a joint decision-making scheme is generated, which includes fault warning, topology reconfiguration instructions, and scheduling strategy compensation. The terminal voltage and current responses of the supercapacitor bank are collected, and the equivalent series resistance, equivalent parallel capacitance and leakage resistance parameters in the equivalent circuit model are calculated through online identification technology. The calculated parameter values ​​are used as the parameters of the equivalent circuit model updated in real time. Calculate the parameter drift of the real-time updated equivalent circuit model parameters relative to the initial calibration value, extract the drift rate and drift acceleration based on the time series features of the parameter drift, and dynamically generate an adaptive threshold according to the coupling relationship between the drift rate and the drift acceleration; When the parameter drift exceeds the adaptive threshold, the contribution ratio of each model parameter in the parameter drift is analyzed, the dominant degradation parameter is identified, and fault warning information is generated based on the physical meaning of the dominant degradation parameter. Based on the fault warning information, the location of the degraded individual cell in the supercapacitor bank is located, a topology reconfiguration instruction is generated to isolate the individual cell location and reconfigure the electrical connection path, the power output capacity attenuation coefficient and the energy reserve attenuation coefficient are calculated based on the equivalent parameters of the reconfigured topology, and a scheduling strategy compensation scheme is generated based on the power output capacity attenuation coefficient and the energy reserve attenuation coefficient. The fault warning information, the topology reconfiguration instruction, and the scheduling strategy compensation scheme are combined to form a joint decision-making scheme.

7. The method according to claim 6, characterized in that, The equivalent series resistance, equivalent parallel capacitance, and leakage resistance parameters in the equivalent circuit model are calculated using online identification technology. The calculated parameter values ​​are then used as the parameters of the equivalent circuit model in real time. The amplitude of the terminal voltage jump and the current step response of the supercapacitor bank during the charging and discharging transition are collected, and the current resistance value of the equivalent series resistance is calculated based on the ratio of the amplitude of the terminal voltage jump to the current step response. The terminal voltage rise curve of the supercapacitor bank during the constant current charging stage is collected. The voltage change rate sequence is obtained by performing time-domain differentiation on the terminal voltage rise curve. The real-time capacitance value of the equivalent parallel capacitor is inverted based on the numerical relationship between the voltage change rate sequence and the charging current. The self-discharge decay curve of the terminal voltage of the supercapacitor bank during the resting stage is collected. The decay time constant is extracted by exponential fitting of the terminal voltage self-discharge decay curve. The current value of the leakage resistance is calculated based on the decay time constant and the real-time capacitance value. The current resistance value of the equivalent series resistor, the real-time capacitance value of the equivalent parallel capacitor, and the current resistance value of the leakage resistor are timestamped to establish a parameter update record sequence. The parameter values ​​of multiple consecutive sampling periods in the parameter update record sequence are smoothed by a sliding window to eliminate the interference of transient noise on parameter identification. The filtered parameter values ​​are used as the parameters of the equivalent circuit model after real-time update.

8. A supercapacitor-based intelligent control system for oil well fracturing power supply, used to implement the method as described in any one of claims 1-7, characterized in that, include: The acquisition and prediction unit is used to acquire multi-source heterogeneous data from the fracturing operation site, perform multi-scale time-frequency decomposition on the multi-source heterogeneous data, extract the transient power impact characteristics and steady-state energy consumption characteristics of the fracturing operation, construct a nonlinear dynamic model of the fracturing load, and predict the power demand evolution trajectory of multiple future time windows based on the nonlinear dynamic model. The energy scheduling unit is used to establish a dynamic mapping relationship between the state of charge of the supercapacitor bank and the fracturing operation sequence based on the power demand evolution trajectory. By solving a multi-objective optimization problem that includes energy constraints, power constraints and lifetime constraints, it generates a globally optimal energy scheduling trajectory that covers pre-charging timing, discharge depth allocation and main power supply coordination ratio. During the execution process, the globally optimal energy scheduling trajectory is rolled over and corrected. The power distribution unit is used to control the energy interaction between the supercapacitor bank and the main power supply through the adaptive power distribution unit according to the global optimal energy scheduling trajectory, so as to achieve precise power supply to the fracturing equipment. The identification feedback unit is used to update the equivalent circuit model parameters of the supercapacitor bank in real time through online identification technology. When the drift of the model parameters is detected to exceed the adaptive threshold, a joint decision scheme including fault warning, topology reconfiguration instructions and scheduling strategy compensation is generated, and the execution result of the joint decision scheme is fed back to the nonlinear dynamic model to form a closed-loop optimization.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.