A Multi-Point Servo Adaptive Leveling Method for Electrochemical Energy Storage Power Station Foundation
By using a multi-point servo adaptive leveling method to sense and predict the settlement of the foundation of a lithium iron phosphate energy storage power station in real time, and using extended Kalman filtering and hydraulic servo jacks for dynamic adjustment, the problem of differential settlement of the foundation under extreme geological scenarios is solved, and efficient and reliable foundation leveling and structural protection are achieved.
Patent Information
- Application Number
- CN202611235671.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies lack a method to detect and adaptively compensate for non-uniform settlement of the foundation of lithium iron phosphate energy storage power stations in real time. In particular, it is impossible to effectively adjust differential settlement of the foundation in extreme geological scenarios, which leads to structural damage and economic losses to the power station.
A multi-point servo adaptive leveling method is adopted, which uses MEMS hydrostatic level and auxiliary sensors to sense settlement in real time, combines extended Kalman filter algorithm to predict settlement trend, realizes feedforward compensation and PID feedback control, uses hydraulic servo jack for dynamic adjustment, and introduces adjacent constraint protection mechanism to build a closed loop system of perception-prediction-control-execution.
It enables real-time, high-precision leveling of the foundation of energy storage power stations, shortens the construction cycle, enhances the structure's disaster resistance, reduces economic losses, and maintains long-term reliable operation in extreme environments.
Smart Images

Figure CN123064207A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of electrochemical energy storage power station infrastructure construction technology, specifically involving an adaptive servo leveling control method for non-uniform settlement of the foundation of a lithium iron phosphate energy storage power station under extreme geological scenarios such as the boundary between mountain slopes, loose aeolian sandy land in deserts, deep soft soil layers in coastal mudflats, and high-altitude freeze-thaw sensitive areas. Background Technology
[0002] Lithium iron phosphate energy storage power stations, as key supporting facilities for new power systems, are being rapidly deployed in remote areas and extreme geological scenarios. These extreme geological scenarios share a common core engineering challenge—non-uniform foundation settlement (also referred to as "differential settlement" in this invention). Specifically, this manifests as follows: in mountainous areas, the settlement on the fill side is significantly greater than that on the cut side due to differences in compaction (differential settlement can reach 50–200 mm); soft soil in coastal tidal flats undergoes continuous consolidation settlement under the long-term load of energy storage power stations (settlement rate can reach 5–30 mm / year), with extremely uneven spatial distribution; aeolian sandy lands in deserts experience localized erosion and differential compaction settlement under wind erosion and vibration loads; and freeze-thaw cycles in high-altitude areas cause repeated foundation uplift and subsidence.
[0003] Lithium iron phosphate energy storage power stations have strict requirements for foundation flatness. The prefabricated battery compartment typically adopts a containerized integrated structure, with internal battery modules connected in series / parallel via busbars, and densely packed battery management system (BMS) wiring harnesses. Differential foundation settlement will lead to: tilting and twisting of the prefabricated compartment, additional stress on container corner fittings and frames, and long-term accumulation leading to structural fatigue cracking; stress concentration at the busbar connections between battery modules due to relative displacement, which in extreme cases may cause loose connections, increased contact resistance, localized overheating, or even arcing; uneven electrolyte distribution due to battery cabinet tilting (for some liquid electrolyte batteries), affecting cell consistency and cycle life; and leaks or cracks at the interfaces of auxiliary facilities such as fire-fighting pipelines and air conditioning ducts due to displacement.
[0004] Currently, the foundation engineering of energy storage power stations generally follows the traditional technical approach used in the civil building sector: Rigid foundation schemes (reinforced concrete raft foundations, independent spread foundations): rely on the rigidity of the foundation itself to "harden" resist differential settlement. Once the foundation experiences excessive uneven settlement, the foundation will tilt irreversibly. It can only be remedied by grouting or leveling with shims afterward. This requires the power station to be shut down, has a long construction period (15-30 days), and cannot prevent further settlement.
[0005] Pile foundation schemes (precast pipe piles, cast-in-place piles): Although the load can be transferred to the deep bearing layer, in soft soil areas, the pile foundation itself may still experience differential settlement (fluctuation of the bearing layer at the pile tip, difference in negative skin friction on the pile side, etc.), and the pile top and the pile cap are rigidly connected with no room for adjustment.
[0006] The common drawback of both of these approaches is that they are both "passive rigid bearing" modes, lacking the ability to adaptively adjust to continuous non-uniform settlement. Once differential settlement occurs, remedial measures can only be taken afterward, and the remedial process requires shutdown, resulting in economic losses.
[0007] In addition, although there are active vibration isolation / leveling platforms for precision equipment in the existing technology (such as air-float leveling of optical platforms and CNC machine tool foundation leveling), they are usually designed for indoor constant temperature and humidity environments and micron-level precision requirements. The systems are complex, costly, and dependent on continuous external energy supply, and are not suitable for outdoor energy storage power station scenarios with large spans, large loads (10 to 50 tons per compartment) and long-term maintenance-free operation.
[0008] In summary, the existing technology lacks a foundation leveling method specifically designed for energy storage power station foundations that can sense non-uniform settlement in real time and adaptively compensate for it with servo, is suitable for harsh outdoor environments, and has a failure safety protection mechanism. Summary of the Invention
[0009] (a) Technical problems to be solved This invention aims to solve the core technical problem of the inability to adaptively compensate for non-uniform settlement of the foundation of lithium iron phosphate energy storage power stations in real time under extreme geological scenarios. It provides a basic multi-point servo adaptive leveling method with closed-loop control of "settlement perception → settlement trend prediction → servo actuation → dynamic leveling → adjacent constraint protection".
[0010] Specific objectives include: achieving real-time, high-precision sensing of differential settlement at various support points of the energy storage power station foundation; predicting future settlement trends based on multi-sensor data fusion to achieve feedforward control that "compensates before settlement occurs"; dynamically and independently adjusting the height of each support point through multi-point independent servo actuation to maintain the flatness of the bottom of the battery prefabricated compartment within the allowable range; introducing differential settlement constraints between adjacent points during the leveling process to avoid additional damage to the superstructure caused by the leveling action itself; and possessing safety protection mechanisms such as sensor fault redundancy and hydraulic failure locking to ensure that the system does not cause secondary disasters under abnormal operating conditions.
[0011] (II) Technical Solution The basic multi-point servo adaptive leveling method proposed in this invention consists of a four-layer architecture: a sensing layer, a prediction layer, a control layer, and an execution layer. The overall logical relationship of the system architecture is as follows: The sensing layer is deployed at each support point and in the foundation, collecting real-time raw data on differential settlement and soil conditions. After preprocessing, this data is transmitted to the prediction layer. The prediction layer runs an extended Kalman filter (EKF) algorithm, fusing multi-source data to estimate the current settlement and settlement rate at each point in real time, and extrapolating the predicted settlement at future times. The current estimate and the predicted value are then transmitted to the control layer. The control layer calculates the feedforward compensation based on the predicted value and the PID feedback compensation based on the real-time error. After coordination with adjacent constraints, the final control command is synthesized and sent to the execution layer. The execution layer converts the control command into the displacement action of each hydraulic servo jack. Built-in displacement sensors provide real-time feedback of the actual displacement, forming an inner-loop position closed loop. Simultaneously, the actual height data is transmitted back to the sensing layer for reference benchmark calculation in the next cycle, constituting a complete data closed loop. The specific technical solutions for each layer are as follows.
[0012] 1. Sensing Layer: Multi-point Settlement Sensing System (1) Sensor layout scheme In the foundation of an energy storage power station, each independent prefabricated battery compartment is equipped with a bottom... Support points (in this invention, "support point" and "measuring point" are the same concept and can be used interchangeably) The typical value for the number of levels is 4 or 6; generally, it should not exceed 8 to avoid excessive system complexity. One MEMS static level is installed at each support point. All MEMS static levels are connected to a common reference level tank via liquid connecting pipes. The reference level tank is installed at the most relatively stable location on the entire site (usually the excavation side with the best geological conditions or the deepest pile foundation). For [the following is unclear due to incomplete text]... An energy storage power station composed of prefabricated battery modules, and a total number of MEMS hydrostatic levels. (Including the reference level tank).
[0013] In addition, auxiliary sensors are selectively arranged at the pile foundation location or in the soil layer below the foundation corresponding to each support point: moisture content sensor and pore water pressure gauge (vibrating wire type). The arrangement density of auxiliary sensors is determined according to the complexity of geological conditions: when the geological conditions are uniform, one set is arranged for every 4 to 6 support points; when the geological conditions vary drastically, each support point is equipped with one set.
[0014] (2) Measurement principle of MEMS hydrostatic level Each support point ( The MEMS hydrostatic level measures the instantaneous height of its mounting base relative to the liquid level inside the connecting pipe. (Unit: mm). When the support point Settlement When (unit: mm), the reading of the MEMS hydrostatic level at that point decreases accordingly. Any two support points and The differential settlement is obtained from the difference in readings:
[0015] in During system initialization The initial reading of the MEMS hydrostatic level at the support point (stored in the controller as zero offset). This is the instantaneous reading at the current moment.
[0016] The ultimate goal of system leveling is to bring all support points to the same horizontal plane. Define the reference height. (Unit: mm) represents the median of the real-time height of all support points:
[0017] in For the first The current actual height of the support point (in mm) is given by the following formula:
[0018] That is, the current actual height is equal to the current instantaneous reading (since the initial reading has been deducted as a zero-point offset, see the difference processing in Equation 1; the actual height is the relative value after subtracting the initial offset from the current reading of the MEMS hydrostatic level). The leveling target for each support point is:
[0019] Key technical requirements for MEMS static level: resolution ≤ 0.01 mm, measurement accuracy ± 0.05 mm, measuring range ≥ 200 mm, operating temperature -40℃~+85℃, protection level IP68.
[0020] (3) Data acquisition and preprocessing All sensors are connected to the field embedded controller via an RS-485 industrial bus or a LoRa wireless node. The system defines the following time parameters: MEMS hydrostatic level sampling cycle s, which is also the outer loop leveling control cycle; Auxiliary sensor sampling period s; Hydraulic inner ring position servo control cycle ms; EKF discretization step size s; Forward period of settlement prediction s.
[0021] The unified threshold for determining communication timeout is as follows: if the sensor does not respond for three consecutive sampling cycles, it is determined to be a communication timeout; if the communication interruption of the hydraulic inner ring exceeds 500ms, the actuator alarm will be triggered.
[0022] Data preprocessing includes: median filtering (window width 5) to eliminate impulse noise; moving average (window width 20) to smooth measurement noise; and outlier detection and removal based on the 3σ criterion, where σ is the real-time standard deviation of the data within the current sliding window (not a fixed preset value) to adaptively track changes in measurement noise. Data points exceeding 3 times the real-time standard deviation are removed and replaced with the previous valid value.
[0023] 2. Prediction Layer: EKF-based Multi-Sensor Fusion Settlement Prediction This module estimates the settlement amount and settlement rate of each support point in real time and predicts the settlement amount at future moments, providing a basis for feedforward compensation.
[0024] (1) Definition of state vector and system modeling For a single support point Define the local state vector ,in This represents the cumulative settlement at that point (unit: mm). Settlement rate (unit: mm / day). A two-dimensional EKF estimator is run independently for each support point. The equivalent compression modulus of the soil layer is uniform across the entire site. (Unit: MPa) and equivalent permeability coefficient (Unit: m / s) The global parameter estimator updates the parameters periodically (every 24 hours; however, when a geological abrupt event is triggered—such as a cumulative rainfall exceeding 50 mm within 24 hours or a sudden increase in settlement rate at adjacent measuring points exceeding three times the historical average—then global parameter re-estimation is immediately initiated, without being subject to the 24-hour timer period), and distributes the updated parameters to the EKF at each point as model parameters. The thickness of the compression layer at each support point. (Unit: m) Determined by the geological survey report, it is a local fixed parameter.
[0025] Based on Terzaghi's one-dimensional consolidation theory, the settlement-time relationship of saturated soft soil foundation under constant load is approximately described as follows:
[0026] in The final settlement amount is expressed in mm. The consolidation time constant (in days) of the compressible layer below the measuring point is given by... The calculations are as follows. For geological scenarios where consolidation is not the dominant factor (such as gravelly soil in fill areas), the model automatically corrects the state prediction bias through real-time measurement updates from the EKF. For non-consolidation-dominated scenarios such as freeze-thaw cycles or wind erosion, the system uses real-time measurement updates from a MEMS hydrostatic level as the main basis for state correction, while monitoring abnormal fluctuations in settlement rate as a scene feature identification signal, but the model structure itself remains unchanged.
[0027] Discretize the continuous model and take the sampling interval. s (approximately 1.157 × 10⁻⁶) -5 (day), thus obtaining the state transition equation:
[0028] In the formula: For the first The support point is at the first Cumulative settlement (mm) over one control cycle; For the first The support point is at the first Settlement rate (mm / day) for each control cycle; The EKF discretization step size (day) is given. s; For the first Consolidation time constant (day) of the support point; For the first The support point is located at the first The effective stress change rate (unit: kPa / cycle, i.e., the average rate of change in each control cycle, not the cumulative increment) is determined by the load change and pore water pressure change at the measuring point. Settlement rate-stress gain coefficient (Note: because) Defined as per control cycle ( The rate of change within ) therefore The numerical calculation implicitly includes a time conversion factor. (Ensure the product dimension is mm / day). The process noise follows a zero-mean Gaussian distribution, and its covariance matrix is... .
[0029] Written in matrix form:
[0030] The state transition matrix and the control input matrix are as follows:
[0031] The observation equation integrates information from three types of sensors. For support points with auxiliary sensors, the observation vector... The components are as follows: Direct observation of settlement using MEMS hydrostatic level:
[0032] There is an empirical correlation between soil moisture content and settlement rate (increased moisture → soil softening → accelerated settlement):
[0033] in The initial moisture content (dimensionless, obtained through geological exploration and geotechnical testing). It is the moisture content-settlement rate coupling coefficient (dimensionless, obtained through indoor geotechnical test calibration or geological survey data regression). To observe noise.
[0034] Relationship between pore water pressure and effective stress rate of change (based on Terzaghi's effective stress principle):
[0035] in The initial pore water pressure (unit: kPa, obtained through geological static penetration test or pore pressure dissipation test). This represents the total stress increment (unit: kPa, calculated from the change in the upper load). To observe noise.
[0036] For support points without auxiliary sensors, the observation vector retains only the MEMS hydrostatic level component, i.e. EKF algorithm for automatic dimensionality reduction (observation matrix) (Only the first line is taken). The observation equation is uniformly written as:
[0037] in The observation matrix (its dimensions are dynamically adjusted based on sensor configuration). The known offset term has components that correspond one-to-one with the constant terms in equations 10 to 12 (for measuring points with auxiliary sensors, the MEMS hydrostatic level offset is 0, and the moisture content observation offset is...). , pore pressure observation offset For measurement points without auxiliary sensors, ), For observation noise, the covariance matrix is: .
[0038] (2) EKF recursive algorithm Initialization: Before use The initial settlement rate estimate was obtained by using MEMS hydrostatic level data from 100 sampling periods (i.e., 100 s) through linear regression. Initial state estimation Initial covariance matrix ,in mm (standard deviation of settlement). mm / day (standard deviation of settling rate).
[0039] Prediction step (from) Time's up time):
[0040] Update step (using) (Time observation correction) Kalman gain:
[0041] Status Update:
[0042] Covariance update:
[0043] in It is an identity matrix, and its dimensions are the same as those of the state vector (2×2).
[0044] (3) Multi-step settlement prediction Based on the current EKF state estimate Forward prediction Settlement after time:
[0045] Among them, the prediction step size s (i.e., the predicted settlement after 5 outer ring control cycles), balancing prediction timeliness and accuracy. Predicted value It will be used as the input to the feedforward compensation channel.
[0046] 3. Control Layer: Feedforward-Feedback Composite Adaptive Leveling Control (1) Feedforward compensation channel By utilizing the settlement prediction results from EKF, compensation amounts are output in advance, enabling the system to proactively adjust itself before actual settlement occurs.
[0047] In the formula The median of the predicted settlement for all support points (in mm). The feedforward gain coefficient (dimensionless) has a theoretical optimal value of 1.0, but in practical engineering, it is taken as 0.85 to 0.95 (leaving a margin to avoid overcompensation).
[0048] (2) Feedback PID channel Accurate compensation is performed based on the real-time leveling error of each support point to eliminate residuals and unmodeled disturbances in the feedforward model. Leveling error is defined as follows:
[0049] in As given by Equation 2, Equation 3 provides the control law, with units in mm. An incremental PID control law is used.
[0050] In the formula (dimensionless) (s) -1 ), (s) represent the proportional, integral, and derivative gains, respectively. The PID parameters for each support point are identified through the transfer function during system initialization using step excitation and automatically determined based on internal model control (IMC) tuning rules. The parameter boundary constraints are: , s -1 , s.
[0051] (3) Comprehensive control quantity No. The overall control displacement of the support point is the superposition of feedforward and feedback:
[0052] The command, after being converted from digital to analog, is output to the servo valve driver, controlling the hydraulic servo jack to extend or retract to the target position. The jack's built-in magnetostrictive displacement sensor provides real-time feedback on the actual displacement. Together with the command value, they form an inner-loop position servo closed loop (PID control, control cycle). ms). Outer loop leveling control cycle s.
[0053] (4) Adjacent constraint coordination and protection mechanism During multi-point independent leveling, if the adjustment of a certain support point differs too much from the adjustment of its adjacent support points, it will cause the upper battery prefabricated cabin frame to experience excessive differential displacement, resulting in additional bending moments and shear forces. Therefore, an adjacent constraint coordination mechanism must be introduced before the control output.
[0054] Define the set of adjacent support point pairs (Based on the planar layout of the support points, adjacency relationships include same-side adjacency and diagonal adjacency). The constraint is: the actual height difference between any pair of adjacent support points must not exceed a preset threshold. :
[0055] threshold ,in Battery prefabricated compartment span (unit: mm, typical value) mm), that is mm, conservative design mm.
[0056] In each outer-loop control cycle, check whether the integrated control quantity satisfies the constraints. If not, obtain the corrected control increment by solving a constrained quadratic programming (QP) problem. The objective function of this QP problem is to minimize the weighted sum of squares of the control quantity correction magnitude:
[0057] The constraints are:
[0058] Among them, weight (Dimensionless) is inversely proportional to the distance from the support point to the geometric center of the prefabricated module, with the weight of the center measuring point being greater than that of the corner measuring points (typical value: corner point). Midpoint of the side center point ); mm is the maximum allowable correction amount per cycle, and half of the threshold is taken to ensure a smooth adjustment process. mm represents the maximum stroke of a single side of the hydraulic servo jack.
[0059] The QP problem described above is a standard convex quadratic programming problem with variable dimensions. The solution is obtained by using the interior point method or the effective set method. The computation time on the embedded controller is ≤50ms, which meets the outer loop control cycle requirements.
[0060] If extreme operating conditions render QP unsolvable (e.g., initial elevation difference between adjacent measuring points exceeds limits and adjustment capacity is insufficient), the system executes the following progressive handling strategy: Upon initial unsolvability, the current control quantity remains unchanged and a pre-alarm is triggered (the local controller records the event); upon unsolvability for three consecutive outer loop cycles, an audible and visual alarm is triggered and the remote monitoring platform is notified (alarm information is simultaneously pushed to the local audible and visual alarm, the remote monitoring platform database, and the maintenance personnel's mobile app, achieving three-level synchronous push), and the system is switched to "safety lockout mode" (all jacks maintain their current positions); upon unsolvability for ten consecutive cycles, an emergency shutdown request is automatically reported, requiring maintenance personnel to intervene. After the shutdown request is issued, a power limiting command is automatically sent to the energy storage power station's energy management system (EMS), requiring the charging and discharging power to be reduced to below 30% of the rated value within 15 minutes to reduce the additional impact of cabin load changes on the foundation. The corrected final control quantity is:
[0061] In addition, the system also sets an upper limit constraint on the overall tilt angle of the prefabricated cabin—when the overall tilt angle of any axis of the cabin exceeds... When the angle is approximately 0.11° (as a supplementary safety margin to adjacent constraints), a level one alarm is triggered and a forced amplitude limit is imposed to avoid the inability to limit unidirectional overall tilting by relying solely on adjacent height difference constraints.
[0062] 4. Execution layer: Servo hydraulic actuation system Each support point is equipped with one independently controlled hydraulic servo jack, forming a multi-point independent execution array. The single-point execution unit consists of the following components from top to bottom: 1. Ball joint connection: Installed between the bottom of the precast foundation and the top of the piston rod of the hydraulic servo jack, allowing a rotation angle of ±3°. This releases the bending moment constraint caused by the tilt of the foundation, ensuring that the jack only bears axial force. The ball joint is made of tempered 40Cr alloy steel, with a PTFE self-lubricating spherical gasket, and is designed for maintenance-free operation. An internal rotation angle limit switch is installed in the ball joint; when the rotation angle exceeds ±3°, an independent limit alarm is triggered and reported to the controller.
[0063] 2. Hydraulic Servo Jack: Double-acting hydraulic cylinder, piston stroke ±150mm (total stroke 300mm), rated load capacity selected according to design load classification: 100kN class for single compartment weight ≤15 tons, 200kN class for 15~30 tons, and 400kN class for 30~50 tons. Built-in magnetostrictive displacement sensor (1μm resolution, ±0.01%FS accuracy, sensor stroke 150mm) provides real-time feedback on the actual displacement of the piston rod. The jack is also equipped with mechanical hard limit switches for upper and lower strokes (located at +145mm and -145mm respectively, with a 5mm safety buffer zone). When the piston reaches the hard limit, an emergency stop is triggered and the current state is locked.
[0064] The MEMS hydrostatic level (accuracy ±0.05mm) and the magnetostrictive displacement sensor (accuracy ±0.015mm, sensor stroke 150mm) constitute a dual-layer displacement measurement system. The MEMS hydrostatic level is used for outer ring settlement sensing, and the magnetostrictive sensor is used for inner ring position closing. The two have different accuracy levels and separate functions, so they do not need to be directly integrated. The steady-state deviation of the MEMS hydrostatic level is automatically eliminated by the integral action of the outer ring PID.
[0065] 3. Electro-hydraulic servo valve: High-frequency response proportional servo valve (-3dB frequency response ≥30Hz), directly integrated into the jack cylinder body, shortening the hydraulic pipeline to improve response speed.
[0066] 4. Hydraulic locking valve: A pilot-operated hydraulic check valve installed in the rodless chamber oil circuit of the jack. It automatically closes when the servo valve control signal returns to zero or the system loses pressure (normally closed when power is off), locking the jack in the current position and achieving fail-safe protection.
[0067] 5. Bottom flange connection: The bottom of the jack cylinder is rigidly connected to the top flange of the underground pile foundation (helical pile or micro pile) through a high-strength bolt flange.
[0068] The entire hydraulic system adopts a centralized pump station + distributed servo valve architecture: it is equipped with two miniature hydraulic pump stations (one for operation and one for standby, each with a power ≤ 5kW and a rated pressure of 21MPa). The switching trigger condition between the main pump and the standby pump is: the main pump outlet pressure continuously falls below the safety threshold. If the pressure exceeds 2 MPa for more than 2 seconds, or if a main pump motor fault signal is triggered, the standby pump will... It automatically starts operating within seconds. The system is also equipped with an accumulator group (capacity ≥10L, used to smooth pressure fluctuations and provide emergency power source), stainless steel seamless steel pipe main pipeline + flexible high pressure hose branch (filtration accuracy ≤10μm), return oil filter and air-cooled radiator (heat dissipation power ≥2kW, ensuring that the oil temperature does not exceed +65℃ when the ambient temperature is +50℃).
[0069] 5. Security Protection Mechanism (1) Sensor fault redundancy When the MEMS hydrostatic level at a certain support point is determined to be faulty (judgment condition: data exceeding the 3σ threshold range for 3 consecutive sampling cycles or communication timeout – the unified threshold for communication timeout is no response for 3 consecutive sampling cycles), the settlement estimate of that support point is obtained by inverse distance weighted interpolation of the EKF estimates of adjacent support points:
[0070] in As a support point The set of all adjacent support points, Adjacent support points Estimated EKF settlement (unit: mm). As a support point and The planar distance (unit: m, measured from the basic plan layout).
[0071] When more than half (including half) of the adjacent measuring points of the same support point fail simultaneously, the system determines that the sensing network in that area has failed, downgrades to "independent single-point feedback control mode" (only based on the historical data of the MEMS hydrostatic level of that point for first-order linear extrapolation, no longer relying on neighboring point interpolation), and triggers an area alarm.
[0072] When an auxiliary sensor (water content sensor or pore water pressure gauge) fails, the EKF observation equation automatically removes the component corresponding to that sensor (dimensionality reduction processing) and only uses the remaining effective sensors for state estimation, which also triggers the corresponding level of maintenance alarm.
[0073] (2) Hydraulic failure lock When the hydraulic system pressure is detected to be below the safety threshold ( When the pressure exceeds 100 MPa or the communication between the controller and the actuator is interrupted for more than 500 ms, all hydraulic locking valves will automatically close (normally closed type when power is off), and the jacks will be locked in the current position to ensure that the foundation will not suddenly sink due to pressure loss. At the same time, audible and visual alarms and remote alarms will be triggered.
[0074] (3) Handling control-side faults When the EKF algorithm exhibits numerical divergence (state covariance matrix) The occurrence of non-positive definite elements or diagonal elements exceeding the initial value. When the displacement exceeds 5 mm, the EKF of the corresponding support point is automatically reset and reinitialized (using the MEMS hydrostatic level data from the most recent 60 cycles for re-reset). When the controller detects that it has crashed (watchdog timeout) or the servo valve is stuck (the deviation between the command value and the actual displacement exceeds 5 mm and remains unchanged for 5 seconds), it automatically triggers a system-level safety lock, all jacks maintain their current position, and reports to the remote monitoring platform requesting manual intervention.
[0075] (4) Temperature drift compensation The zero point of the MEMS static level may drift with temperature changes within a temperature range of -40℃ to +85℃. During the system initialization phase, the ambient temperature... Record the zero-point reading at the initial calibration temperature, and measure the real-time ambient temperature using a temperature sensor during operation. According to the linear correction model Zero-point drift correction is performed, among which Temperature drift coefficient (provided by the sensor factory calibration, typical value ≤0.01mm / ℃).
[0076] (5) System initialization and real-time operation process Initialization process (executed after system first power-on or reset): Step I1: Sensor self-test -- All MEMS hydrostatic level, auxiliary sensors, and displacement sensors are powered on and self-tested in sequence to confirm that communication is normal and readings are within a reasonable range.
[0077] Step I2: Reference Plane Calibration – Use a laser level to measure the initial absolute elevation of the upper surface of each support point's platform and record it. (Unit: mm), and this value is stored in the controller as the initial zero-point offset (instead of physically zeroing the sensor, the original sensor reading). (It was fully preserved and used as a calculation benchmark).
[0078] Step I3: EKF Initialization -- Before Data Acquisition MEMS hydrostatic level data for one outer loop period (100s) were used to obtain the initial settlement rate estimate of each support point through linear regression, and the EKF state vector and covariance matrix were initialized.
[0079] Step I4: PID parameter self-tuning -- Apply a 1mm step excitation to each execution channel sequentially, collect displacement step response data (sampling period 10ms, recording time 5s), and fit it into a first-order inertial + pure time-delay model using the least squares method. The PID parameters are calculated based on the IMC tuning rules, and it is verified whether they meet the requirements. , s -1 , If the boundary constraints of s are not satisfied, the default safety parameters will be used. .
[0080] Step I5: Adjacent Constraint Verification and Initial Leveling -- Check whether the elevation difference between adjacent support points meets the requirements in the initial state. Constraints. If the initial elevation difference exceeds the limit, the system prioritizes leveling to eliminate the initial difference. During this stage, the feedforward channel is not enabled (only feedback control + adjacent constraints), and the leveling rate is limited to 0.2 mm / s. When the elevation difference of all adjacent points... After 10 consecutive stable cycles, the initial leveling is determined to be complete, and the system automatically switches to normal operating mode.
[0081] Step I6: Enter normal operation mode and enable feedforward + feedback composite control.
[0082] Real-time operation process (each outer loop control cycle) s execution): Step R1: Collect the current readings of all MEMS hydrostatic level, auxiliary sensors, and jack displacement sensors.
[0083] Step R2: Perform median filtering (window 5) + moving average (window 20) + outlier removal (real-time 3σ criterion) on the raw readings of the MEMS hydrostatic level.
[0084] Step R3: Each support point independently performs EKF prediction and update to obtain... and .
[0085] Step R4: Calculate the forward predicted settlement ( s).
[0086] Step R5: Calculate the reference height .
[0087] Step R6: Calculate the feedforward compensation amount .
[0088] Step R7: Calculate the leveling error and feedback compensation amount .
[0089] Step R8: Synthesize the overall control quantity .
[0090] Step R9: Check adjacent constraints and overall tilt angle constraints. If all adjacent pairs satisfy the constraints, output directly. Otherwise, solving the QP problem yields... Output .
[0091] Step R10: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] The voltage commands are converted into voltage commands for each servo valve, which drive the jacks to perform displacement.
[0092] Step R11: Jack inner ring position servo closed loop (PID, (ms) Ensures that the actual displacement is tracked by the command value.
[0093] Step R12: Record the operational data for this cycle (settlement amount, settlement rate, compensation amount, error, etc. of each support point), store it in the local database, and upload it to the remote monitoring platform through the IoT gateway.
[0094] Step R13: Wait for the next control cycle.
[0095] (III) Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects: 1. This invention represents a fundamental shift from "passive rigid bearing" to "active flexible self-adaptation." Traditional rigid foundations experience irreversible differential settlement once it occurs. This invention, through closed-loop servo control, enables the foundation to possess intelligent adjustment capabilities of "sensing-decision-execution," introducing active control technology into the field of energy storage power station foundation engineering for the first time.
[0096] 2. Real-time and predictive capabilities of non-uniform settlement compensation. Based on a composite control architecture combining EKF multi-sensor fusion for settlement prediction feedforward and incremental PID feedback, a real-time closed-loop system of "simultaneous settlement, prediction, and compensation" is achieved, with a leveling response time ≤5s and steady-state flatness maintained within ±2mm. Compared to traditional post-treatment grouting / shim remedial solutions (requiring 15-30 days of downtime), the economic benefits are significant.
[0097] 3. Significantly shortened construction period. The foundation system of this invention does not require large-scale excavation, formwork, concrete pouring, and curing. It adopts a dry and rapid construction scheme of spiral pile driving, precast foundation assembly, and jack flange bolt connection. The foundation construction period is shortened from 30-60 days in the traditional scheme to 5-12 days, a reduction of more than 80%.
[0098] 4. Significantly improved structural disaster resistance. By eliminating secondary internal forces caused by differential settlement, the additional stress of the prefabricated battery compartment frame is reduced by more than 70%; the adjacent constraint coordination mechanism ensures that the leveling process does not cause secondary damage to the structure; the hydraulic failure locking mechanism ensures safety under extreme working conditions, and the overall assessment shows that the structure's resistance to settlement disasters is improved by more than 50%.
[0099] 5. Full lifecycle maintainability and scalability. Each servo actuator can be independently repaired or replaced without affecting the normal operation of other units; the system supports continuous adaptive adjustment during the operation period, covering the entire lifecycle of the energy storage power station (20-30 years); the algorithm and hardware architecture can be scaled to energy storage power stations of any size and layout.
[0100] 6. Robustness to extreme environments. The MEMS hydrostatic level (IP68 protection, operating temperature -40 to +85℃), industrial-grade embedded controller, and hydraulic locking valve fail-safe design ensure long-term reliable operation of the system in extreme environments such as mountains, deserts, coastal mudflats, and high altitudes. Attached Figure Description
[0101] Figure 1 This is a block diagram of the overall system architecture of the present invention. Figure 2 EKF Multi-Sensor Fusion Settlement Prediction Flowchart Figure 3 Overall flowchart of the adaptive leveling control method Figure 4 Schematic diagram of a single-support-point execution unit structure Detailed Implementation
[0102] The present invention will be further described in detail below with reference to specific embodiments. This embodiment takes a lithium iron phosphate energy storage power station built in a coastal area with a thick soft soil layer as a typical application scenario to fully illustrate the technical implementation path of the present invention. This solution is also applicable to the foundation leveling project of electrochemical energy storage power stations in other extreme geological scenarios such as the junction of mountain slopes, loose aeolian sandy land in deserts, and high-altitude freeze-thaw sensitive areas. The scope of protection is not limited to the scenario of this embodiment.
[0103] In this embodiment, the rated total weight of a single standard battery prefabricated compartment is: tons, cabin span Single-compartment bottom setting Each independent support point ( ), respectively arranged at the four corners of the cabin; a total of ), were configured throughout the station. Prefabricated battery compartment ( ), supporting The total number of MEMS hydrostatic levels at the entire station is [number missing], serving as a public reference level tank. tower.
[0104] The basic multi-point servo adaptive leveling method of this invention adopts a four-layer closed-loop architecture of perception layer, prediction layer, control layer, and execution layer. The overall logic is as follows: The perception layer collects settlement data and foundation soil state parameters of each support point in real time, and transmits them to the prediction layer after preprocessing; The prediction layer fuses multi-source data through the extended Kalman filter (EKF) algorithm, estimates the settlement state of each point in real time and predicts the future settlement trend, and outputs the estimated value and the predicted value to the control layer synchronously; The control layer generates a feedforward compensation amount based on the predicted settlement amount and a PID feedback compensation amount based on the real-time settlement error, and outputs the final control command to the execution layer after coordination and correction by adjacent constraints; The execution layer executes displacement actions through hydraulic servo jacks, and forms an inner-loop position closed loop with built-in displacement sensors, while simultaneously transmitting the actual height data back to the perception layer, thus forming a complete closed-loop control system.
[0105] I. Sensing Layer: Multi-point Settlement Sensing System (1) Sensor layout scheme In each battery prefabrication compartment Each corner support point is installed. Unit MEMS hydrostatic level; all hydrostatic levels are connected via liquid connecting pipes. A public reference liquid level tank is fixed at the top of the deep pile foundation on the excavation side where the geological conditions are most stable, serving as a public benchmark for settlement measurement across the entire site.
[0106] A moisture content sensor and a vibrating wire pore water pressure gauge are auxiliaryly deployed in the foundation soil layer. In this embodiment, the geological conditions are generally homogeneous, and each... Layout of support points The density scheme of auxiliary sensors is as follows: the auxiliary sensors are buried at the middle depth of the compression layer below the support point; in areas with drastic changes in geological conditions, auxiliary sensors can be arranged at each support point.
[0107] (2) Settlement measurement and benchmark calculation The key technical parameters of the MEMS hydrostatic level instrument meet the following requirements: resolution Measurement accuracy Range Operating temperature Protection rating: IP68.
[0108] Let the first Each support point is The original reading of time is The initial reading during system initialization is And it is stored in the controller as the zero-point offset. Any two support points , The differential settlement between them is calculated using the following formula:
[0109] The system uses the median of the real-time actual heights of all support points as the reference height. ,Right now:
[0110] The current actual height of the support point This refers to the relative height value after zero-difference differentiation processing. The leveling target for each support point is to be level with the reference benchmark height, i.e. .
[0111] (3) Data acquisition and preprocessing All sensors are connected to the field embedded controller via an RS-485 industrial bus or a LoRa wireless node. The system time parameters are set as follows: MEMS hydrostatic level sampling cycle It is consistent with the outer ring leveling control cycle; Sampling cycle of auxiliary sensors such as moisture content and pore water pressure gauge ; Hydraulic inner ring position servo control cycle ; EKF discretization step size ; Forward period of settlement prediction .
[0112] Sensor communication fault determination rules: continuous If there is no response within one sampling period, it is determined to be a communication timeout; the hydraulic inner loop communication interruption exceeds [period]. The actuator alarm is triggered.
[0113] The data preprocessing process is executed sequentially: First, a window with a width of... Median filtering was used to eliminate impulse noise; subsequently, a window width of [missing information] was employed. The moving average smooths the measurement noise; finally, based on real-time... The criteria are used for outlier detection and removal, among which... The real-time standard deviation of the data within the current sliding window, exceeding... Data points that are more than twice the real-time standard deviation are discarded and replaced with the previous valid value.
[0114] II. Prediction Layer: EKF-based Multi-Sensor Fusion Settlement Prediction Each support point independently operates a two-dimensional extended Kalman filter estimator to achieve settlement state estimation and trend prediction.
[0115] (1) State vector and system modeling For a single support point Define the local state vector ,in Cumulative settlement (unit: ), Settlement rate (unit: ).
[0116] A settlement prediction model was constructed based on Terzaghi's one-dimensional consolidation theory. The settlement-time relationship for saturated soft soil foundations is as follows:
[0117] in This is the final settlement. For the first The consolidation time constant of the measuring point is determined by... Calculated; The thickness of the compressible layer below this measuring point is determined as a local fixed parameter by the geological survey report; For the equivalent compression modulus of the soil layers throughout the site, The equivalent permeability coefficient is calculated by the global parameter estimator for each... Updated every hour, when continuous Hourly cumulative rainfall exceeded Or the settlement rate of adjacent measuring points suddenly increases by more than When the value is doubled compared to the historical average, a global parameter re-estimation is immediately triggered, without being limited by the timing period.
[0118] Discretizing the continuous model yields the state transition equations:
[0119] Written in matrix form:
[0120] Where the state transition matrix
[0121] The effective stress change rate is determined by the changes in load at the measuring point and the changes in pore water pressure. Settlement rate-stress gain coefficient; The noise is a zero-mean Gaussian process, and the covariance matrix is... .
[0122] For geological scenarios where non-soft soil consolidation is the dominant factor, the model automatically corrects state prediction deviations through real-time EKF measurements. For scenarios such as freeze-thaw cycles and wind erosion, the system uses real-time measurements from a MEMS hydrostatic level as the primary basis for correction, while monitoring abnormal fluctuations in settlement rate as a scenario identification signal, and the model structure remains unchanged.
[0123] (2) Construction of observation equations For support points equipped with auxiliary sensors, the observation vector includes three types of observations: those from a MEMS hydrostatic level, a water content sensor, and a pore water pressure gauge. The observation equations are unified as follows:
[0124] The observed components are as follows: 1. Direct observation of settlement using MEMS hydrostatic level:
[0125] The corresponding observation offset is ; 2. Coupling of water content observation with empirical settling rate:
[0126] in For initial moisture content, The moisture content-settlement rate coupling coefficients are obtained from geotechnical tests. 3. Pore water pressure observation is based on the effective stress principle:
[0127] in The initial pore water pressure, This represents the total stress increment.
[0128] For support points without auxiliary sensors, the observation vector retains only the MEMS hydrostatic level component, and the observation matrix is automatically dimensionality-reduced. For observation noise, the covariance matrix is: .
[0129] (3) Implementation of EKF recursive algorithm Initialization: Before data acquisition after system power-on Each sampling period ( Using MEMS hydrostatic data, the initial settlement rate estimates for each support point are calculated through linear regression, and an initial state vector is constructed. Initial covariance matrix ,in , .
[0130] Prediction step: by Extrapolation of state estimates at time points Prior states and covariance at each time step:
[0131] Update step: Utilize Correcting prior states with time-varying observations, calculating Kalman gain, updating state estimates, and covariance:
[0132] in for Identity matrix.
[0133] (4) Implementation of multi-step settlement prediction Based on the current time State estimates, forward prediction Subsequent settlement:
[0134] The predicted settlement is used as the input to the feedforward compensation channel to achieve pre-compensation before settlement occurs.
[0135] III. Control Layer: Feedforward-Feedback Composite Adaptive Leveling Control (1) Feedforward compensation channel The compensation amount is output in advance using the settlement prediction results of EKF. The feedforward control amount is calculated according to the following formula:
[0136] in The median of the predicted settlement for all support points; the feedforward gain coefficient in this embodiment. At the theoretical optimal value To avoid overcompensation, a margin should be left on the basis; the range of engineering values is as follows: .
[0137] (2) Feedback PID channel Define leveling error The feedback compensation amount is calculated using an incremental PID control law:
[0138] PID parameters are determined through a self-tuning method: during system initialization, parameters are applied to each execution channel. Step excitation was performed, displacement response data were collected, and a first-order inertial + pure time-delay model was fitted using the least squares method. The PID parameters are tuned based on internal model control (IMC) rules. The parameter boundary constraints are: , , If the self-tuning result exceeds the boundary, the default safety parameters will be used. , , .
[0139] (3) Synthesis of comprehensive control quantities No. The total control displacement of each support point is the superposition of feedforward and feedback:
[0140] Control commands are converted from digital to analog and output to the servo valve driver, which in turn drives the hydraulic servo jack. The jack's built-in magnetostrictive displacement sensor provides real-time feedback on the actual displacement, forming an inner-loop position servo closed loop. The inner-loop control cycle is... .
[0141] (4) Adjacent constraint coordination and protection mechanism Define a set of adjacent support point pairs based on the planar layout of the support points. This includes adjacent relationships on the same side and diagonally adjacent relationships. The constraint is that the actual height difference between any two adjacent support points after adjustment must not exceed a preset threshold. :
[0142] threshold Corresponding span The prefabricated cabin has a threshold range of This embodiment adopts a conservative design. .
[0143] Each outer loop control cycle performs constraint verification on the integrated control quantity: if the constraints are met, the output is performed directly; otherwise, the corrected control increment is obtained by solving a constrained quadratic programming (QP) problem. .
[0144] The objective function of the QP problem is to minimize the weighted sum of squares of the control quantity correction magnitudes:
[0145] The constraints include: 1. Adjacent Elevation Difference Constraint: After adjustment, the elevation difference between adjacent points shall not exceed [the specified value]. ; 2. Single-period correction constraint: In this embodiment, The threshold value is set at half to ensure smooth adjustment. 3. Travel constraints: The maximum stroke of the jack on one side .
[0146] Weighting coefficient It is inversely proportional to the distance from the support point to the geometric center of the prefabricated module, with typical values being: corner point Midpoint of the side center point This embodiment The weights of the corner support points are all taken as... The embedded controller uses the interior-point method or the effective set method to solve this convex quadratic programming problem. The computation time per cycle is [not specified]. ,satisfy Outer loop control cycle requirements.
[0147] The revised final control value is:
[0148] If QP has no feasible solution under extreme conditions, a progressive handling strategy is implemented: If no solution is found initially: keep the current control value unchanged, trigger a local pre-alarm and record the event; continuous If the cycle is unsolvable: trigger an audible and visual alarm and push it simultaneously to the remote monitoring platform and the mobile terminal of the maintenance personnel, the system switches to the safety lock mode, and all jacks maintain their current positions; continuous No solution for the current cycle: Automatically report emergency shutdown requests and simultaneously send power limiting commands to the energy storage power station's EMS, requiring... The charging and discharging power will be reduced to the rated value within minutes. the following.
[0149] In addition, supplementary constraints are set for the overall tilt angle of the cabin: when the overall tilt angle of any axis of the cabin exceeds... (about When the angle is 1), a level 1 alarm is triggered and a forced amplitude limit is imposed to avoid the inability to limit the overall tilt in one direction by relying solely on the adjacent height difference constraint.
[0150] IV. Execution Layer: Servo Hydraulic Actuation System Each support point is configured with Each independently controlled hydraulic servo jack forms a multi-point independent execution array.
[0151] (1) Single-point execution unit structure The single-point actuator, from top to bottom, consists of a ball joint connector, a hydraulic servo jack, an electro-hydraulic servo valve, a hydraulic locking valve, and a bottom flange. Specific implementation parameters are as follows: 1. Ball joint connection: Installed between the bottom of the precast foundation and the top of the jack piston rod, allowing... The angle adjustment releases the bending moment caused by the tilt of the pier, ensuring that the jack is only subjected to axial force; it is made of tempered 40Cr alloy steel, with a PTFE self-lubricating spherical gasket, and features a maintenance-free design; it has a built-in angle limit switch, which prevents the angle from exceeding... An independent limit alarm is triggered at certain times.
[0152] 2. Hydraulic servo jack: Double-acting hydraulic cylinder, piston stroke (Total Itinerary) Rated load-bearing capacity is classified according to design load: single compartment weight Ton selection class, Ton selection class, Ton selection Level; This embodiment selects Level. The jack has a built-in magnetostrictive displacement sensor with a resolution of [missing information]. Precision Real-time feedback of piston rod displacement; equipped with mechanical hard limit switches for upper and lower strokes, located at... Reserved space Safety buffer: triggers an emergency stop and locks the device when the hard limit is reached.
[0153] 3. Electro-hydraulic servo valve: High-frequency response proportional servo valve, -3dB frequency response ≥30Hz, directly integrated into the jack cylinder body, shortening the pipeline and improving the response speed.
[0154] 4. Hydraulic locking valve: Pilot-operated hydraulic check valve, installed in the rodless chamber oil circuit of the jack, is normally closed when power is off; it automatically closes when the servo valve signal returns to zero or the system loses pressure, locking the jack position and achieving fail-safe protection.
[0155] 5. Bottom flange connection: The bottom of the jack cylinder is rigidly connected to the top flange of the underground helical pile or micro pile through a high-strength bolt flange.
[0156] MEMS hydrostatic level (outer ring settlement sensing, high accuracy) ) and magnetostrictive displacement sensor (inner loop position closed loop, accuracy) This constitutes a dual-layer measurement system. The two layers are functionally separated and have matching accuracy levels, so there is no need for direct fusion. Steady-state deviations are automatically eliminated through the integral action of the outer loop PID.
[0157] (2) Configuration of the entire station's hydraulic system The entire station adopts a centralized pumping station + distributed servo valve architecture, configured with One miniature hydraulic pump station, one in use and one on standby, single unit power Rated pressure Main / standby pump switching trigger condition: The main pump outlet pressure remains below the safety threshold. Exceed Or triggered by a main pump motor fault signal; standby pump switching time .
[0158] System supporting volume The accumulator array is used to smooth pressure fluctuations and provide emergency power; the main pipeline uses seamless stainless steel pipes, and the branches use flexible high-pressure hoses with high filtration accuracy. Configure oil return filter and heat dissipation power. The air-cooled radiator ensures the ambient temperature. Oil temperature not exceeding .
[0159] V. Implementation of Safety Protection Mechanism (1) Sensor fault redundancy When a certain support point is continuously supported by the MEMS static level Data from each sampling period exceeds If the range or communication times out, it is determined to be a sensor malfunction. After a malfunction, the estimated settlement value of the support point is obtained by inverse distance weighted interpolation of the EKF estimates of the adjacent support points:
[0160] in The set of adjacent support points of the fault point. The distance between two points in a plane is denoted as .
[0161] When more than half of the adjacent measuring points of the same support point fail simultaneously, the sensing network of that area is determined to be faulty, the system is downgraded to an independent single-point feedback control mode, and only first-order linear extrapolation is performed based on the historical data of that point, while triggering an area alarm.
[0162] When the moisture content sensor or pore water pressure gauge fails, the EKF observation equation automatically removes the corresponding component, performs state estimation by reducing the dimension, and triggers a maintenance alarm.
[0163] (2) Hydraulic failure lock When the hydraulic system pressure is detected to be lower than Or the communication between the controller and the actuator is interrupted for more than [time period missing] At this time, all hydraulic locking valves are automatically de-energized and closed, and the jacks are locked in their current positions to prevent the foundation from suddenly sinking; at the same time, on-site audible and visual alarms and remote alarms are triggered.
[0164] (3) Handling control-side faults When the EKF state covariance matrix is non-positive definite, or the diagonal elements exceed the initial values. When the time factor is doubled, it is determined that the algorithm is diverging, the EKF of the corresponding support point is automatically reset, and the nearest value is used. The measurement data for each cycle is reinitialized.
[0165] When the controller watchdog times out, or the servo valve command value deviates from the actual displacement by more than [a certain amount]... And continue When there is no change, a system-level safety lock is automatically triggered, all jacks maintain their current positions, and a request for manual intervention is sent to the remote monitoring platform.
[0166] (4) Temperature drift compensation During system initialization at the calibration temperature The system records the zero-point readings of each MEMS hydrostatic level; during operation, it collects real-time ambient temperature data via a temperature sensor. Zero-point drift linear correction is performed using the following formula:
[0167] Among them, temperature drift coefficient Typical values are provided by the sensor's factory calibration. .
[0168] VI. System Implementation Process (1) Initialization process (executed after first power-on or reset) Step I1: Sensor self-test. All MEMS hydrostatic level, auxiliary sensors, and displacement sensors are powered on and self-tested sequentially to confirm normal communication and that the readings are within a reasonable range.
[0169] Step I2: Reference plane calibration. The initial absolute elevation of the upper surface of the support platform at each support point is measured using a laser level. The initial zero-point offset is stored in the controller, and the original initial readings of the sensor are synchronously and completely preserved.
[0170] Step I3: EKF initialization. Continuous data acquisition. One outer ring cycle ( Using MEMS hydrostatic level data, the initial settlement rate of each support point is calculated through linear regression, and the EKF state vector and covariance matrix are initialized.
[0171] Step I4: PID parameter self-tuning. Apply parameters sequentially to each execution channel. Step excitation, data acquisition Displacement response data are used to fit a first-order inertial + pure time-delay model. PID parameters are tuned based on IMC rules and boundary constraints are checked. If the constraints are not met, default safety parameters are used.
[0172] Step I5: Adjacent Constraint Verification and Initial Leveling. Check the elevation difference between adjacent support points in the initial state. If it exceeds... Initial leveling is performed using only feedback control and adjacent constraints, with the leveling rate limited to [value missing]. When the elevation difference of all adjacent points And it remains stable After one cycle, the initial leveling is determined to be complete.
[0173] Step I6: The system switches to normal operation mode and enables feedforward + feedback composite control.
[0174] (2) Real-time operation process (each outer loop control cycle) (Execute once) Step R1: Collect the current readings of all MEMS hydrostatic level, auxiliary sensors and jack displacement sensors.
[0175] Step R2: Perform median filtering, moving average, and other processing sequentially on the raw data from the MEMS hydrostatic level. Preprocessing for outlier removal.
[0176] Step R3: Each support point independently performs the EKF prediction step and update step to obtain the optimal estimate of the current settlement amount and settlement rate.
[0177] Step R4: Based on the current state estimate, calculate Forward predicted settlement.
[0178] Step R5: Calculate the median of the actual heights of all support points, and use it as the current reference height. Step R6: Calculate the feedforward compensation based on the predicted settlement and the baseline predicted value.
[0179] Step R7: Calculate the leveling error of each support point and calculate the feedback compensation amount using incremental PID.
[0180] Step R8: Superimpose the feedforward and feedback compensation quantities to obtain the comprehensive control quantity.
[0181] Step R9: Verify the adjacent elevation difference constraints and the overall tilt angle constraints. If the constraints are met, directly output the integrated control quantity; otherwise, solve the QP problem to obtain the correction increment and synthesize the final control quantity.
[0182] Step R10: Convert the final control quantity into a servo valve voltage command and output it to the execution layer.
[0183] Step R11: The hydraulic jack's inner ring position servo closed-loop tracks the command displacement, control cycle. Step R12: Record all running data for this cycle, store it in the local database, and upload it to the remote monitoring platform.
[0184] Step R13: Wait for the next control cycle and repeat.
[0185] Implementation effect verification Through the above implementation methods, this method can achieve real-time adaptive compensation for non-uniform settlement of energy storage power station foundations under extreme geological scenarios: leveling response time Steady-state flatness can be maintained at Within [a certain range]; the foundation adopts a dry, rapid construction scheme of helical pile driving + precast pile cap assembly + jack flange connection, which can shorten the construction period to [a certain timeframe]. The time is shorter than that of traditional rigid foundation schemes. In summary, by eliminating the additional stress from differential settlement, the additional stress on the prefabricated battery compartment frame structure is reduced. The above improves the structure's resistance to settlement disasters. The system possesses multiple failure safety protection mechanisms, enabling long-term maintenance-free operation in harsh outdoor environments, covering energy storage power stations. The entire life cycle.
[0186] This solution breaks through the limitations of the traditional energy storage foundation's "passive rigid bearing and post-event remediation" technical approach. It introduces a "sensing-prediction-servo-constraint" closed-loop active control system into the field of energy storage power station foundation engineering, solving the industry problem of the inability to provide real-time adaptive compensation for non-uniform settlement under extreme geological conditions.
Claims
1. A multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station, characterized in that, Includes the following steps: Step 1, Sensing Step: MEMS static level instruments are deployed at each support point of the energy storage power station foundation to collect real-time height data of each support point relative to the liquid level in the common reference liquid level tank, and the collected raw data is preprocessed. Step 2, Prediction Step: Establish a state vector for each support point. The state vector includes the cumulative settlement and settlement rate. Construct a discrete state transition equation based on one-dimensional consolidation theory. Use the extended Kalman filter algorithm to fuse multi-sensor observation data, estimate the current settlement and settlement rate of each support point in real time, and extrapolate to predict the settlement at future times. Step 3, Control Steps: Calculate the feedforward compensation amount based on the predicted settlement amount of each support point, calculate the feedback compensation amount based on the real-time leveling error of each support point, combine the feedforward compensation amount and the feedback compensation amount to synthesize the comprehensive control amount, and output the final control command after performing adjacent constraint coordination correction on the comprehensive control amount. Step 4, Execution Steps: The final control command is converted into the displacement action of the hydraulic servo actuator at each support point, driving each support point to rise and fall independently, while the actual displacement of each actuator is fed back in real time to form a position closed loop; The sensing step, prediction step, control step, and execution step are executed cyclically according to the control cycle. The settlement data of the sensing step is transmitted back to the prediction step in real time for state estimation update in the next cycle, forming a fully closed-loop adaptive leveling control.
2. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, In the sensing step, each independent battery prefabricated compartment is equipped with a bottom One support point, One MEMS static level is installed at each support point; all MEMS static levels are connected to a common reference level tank via liquid connecting pipes, and the reference level tank is installed at the most stable position in the entire field; for those... The energy storage power station consists of prefabricated battery modules, and the total number of MEMS hydrostatic levels is [number missing]. Including the level at the reference liquid level tank; Each support point Current actual height Take the relative value of the current instantaneous reading of the MEMS hydrostatic level after deducting the initial zero-point offset; reference reference height. Defined as the median of the real-time actual height of all support points; the leveling target for each support point is to make the actual height of that support point equal to the reference reference height.
3. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, In the prediction step, for a single support point Define the local state vector in For cumulative settlement, Settlement rate; Discrete state transition equation is in To discretize the step size, This is the consolidation time constant of the support point. For effective stress change rate, Settlement rate-stress gain coefficient, , This is process noise; The observation equation includes at least the settlement component directly observed by the MEMS hydrostatic level. For support points equipped with auxiliary sensors, the observation equation also includes the soil moisture content observation component and the pore water pressure observation component. The extended Kalman filter algorithm is executed recursively in prediction and update steps. The prediction step obtains the state prior estimate and covariance prior estimate based on the state transition equation. The update step obtains the state posterior estimate and covariance posterior estimate based on the observed values.
4. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, In the control steps, the feedforward compensation amount Calculate using the following formula: in For the first Support points predict the future Settlement after a certain time. Predict the median settlement for all support points. This is the feedforward gain coefficient; The feedback compensation is calculated using an incremental PID control law, based on the first... Real-time leveling error of support points get; Comprehensive control quantity .
5. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, The control steps specifically include: Adjacent constraint coordination and correction. Define the set of adjacent support point pairs The constraint is that the actual height difference between any two adjacent support points must not exceed a preset threshold. : In each control cycle, the integrated control quantity is checked to see if it meets the constraints. If not, the control quantity correction for each support point is obtained by solving a constrained quadratic programming problem. The objective function is to minimize the weighted sum of squares of the control quantity correction magnitude, and the constraints are the elevation difference between adjacent points, the maximum correction amount in a single cycle, and the maximum travel at a single point. The final corrected control quantity is then calculated. Output to the execution steps.
6. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 5, characterized in that, The threshold ,in For the span of the prefabricated battery compartment, Values mm; the objective function of the quadratic programming problem is: Among them, weight The constraint is inversely proportional to the distance from the support point to the geometric center of the prefabricated module; the constraint conditions also include as well as This refers to the maximum stroke on one side of the hydraulic servo actuator. When the quadratic programming problem has no feasible solution, the system executes a progressive handling strategy: if there is no solution for the first time, the current control quantity remains unchanged and a pre-alarm is triggered; if there is no solution for 3 consecutive control cycles, an audible and visual alarm is triggered and the remote monitoring platform is notified, while the system is switched to a safety lockout mode; if there is no solution for 10 consecutive cycles, an emergency shutdown request is automatically reported.
7. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, In the execution steps, each support point is equipped with one independently controlled hydraulic servo jack, forming a multi-point independent execution array; the hydraulic servo jack is a double-acting hydraulic cylinder with a built-in magnetostrictive displacement sensor that provides real-time feedback on the actual displacement of the piston rod; the magnetostrictive displacement sensor and the MEMS hydrostatic level form a dual-layer displacement measurement system, wherein the MEMS hydrostatic level is used for outer ring settlement sensing, and the magnetostrictive displacement sensor is used for inner ring position closure; The piston rod of the hydraulic servo jack is connected to the bottom of the precast foundation via a ball joint. The ball joint allows... The hydraulic servo jack has a built-in angle limit switch; the bottom of the cylinder is rigidly connected to the underground pile foundation via a flange; the rodless chamber oil circuit of the hydraulic servo jack is equipped with a normally closed hydraulic locking valve that automatically closes and locks the jack in the current position when the servo valve control signal returns to zero or the system loses pressure.
8. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, When the MEMS hydrostatic level at a certain support point is determined to be faulty, the settlement estimate of that support point is obtained by inverse distance weighted interpolation of the extended Kalman filter settlement estimate of the adjacent support points; when more than half of the adjacent measuring points of the same support point fail at the same time, the system is downgraded to an independent single-point feedback control mode; when the auxiliary sensor fails, the extended Kalman filter observation equation automatically removes the corresponding component of the sensor and performs dimensionality reduction. When the extended Kalman filter algorithm exhibits numerical divergence, the extended Kalman filter at the corresponding support point is automatically reset and reinitialized; when the controller watchdog times out or the servo valve jams, a system-level safety lock is automatically triggered, and all jacks maintain their current positions.
9. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, During operation, the MEMS hydrostatic level uses a temperature sensor to measure the real-time ambient temperature and performs temperature drift correction on the measurement zero point according to a linear correction model. in The initial calibration temperature. This is the temperature drift coefficient.
10. The multi-point servo adaptive leveling method for the foundation of an electrochemical energy storage power station according to claim 1, characterized in that, Upon initial power-on or reset, the system executes an initialization procedure, which includes: Sensor self-test steps: Power on all MEMS hydrostatic level, auxiliary sensors and displacement sensors in sequence to perform self-tests and confirm normal communication; Reference plane calibration steps: Use a laser level to measure the initial absolute elevation of the upper surface of each support point and record it as the initial zero point offset, then store it in the controller; Extended Kalman filter initialization steps: Collect MEMS hydrostatic level data for the first 100 control cycles, obtain the initial settlement rate estimate of each support point through linear regression, and initialize the state vector and covariance matrix of the extended Kalman filter; PID parameter self-tuning steps: Apply step excitation to each execution channel in sequence, collect displacement step response data, fit the transfer function by least squares method and calculate PID parameters based on internal model control rules; Adjacent constraint verification and initial leveling steps: Check whether the elevation difference between adjacent support points in the initial state meets the constraints. If the initial elevation difference exceeds the limit, eliminate the initial difference through feedback control. The leveling rate is limited to [value missing]. mm / s, and switch to normal operation mode after the height difference of all adjacent points meets the constraint and remains stable.
