A Method for Pumping Status Identification and Adaptive Control of High-Strength Concrete in Super High-Rise Buildings

Abstract

This invention discloses a method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings. Addressing industry pain points such as high friction, easily abrupt changes in flow regime, and susceptibility to pipe blockage and segregation in ultra-high-rise C60 and above high-strength concrete pumping, this invention achieves advanced status identification through multi-source time-series data acquisition, specialized feature extraction, and fusion of time-varying friction mechanism and LSTM hybrid model. It employs an MPC and feedforward compensation architecture to achieve multi-level collaborative adaptive control of the main pump, relay pressurization, and admixtures, supporting sensor fault state reconstruction. This method achieves advanced warnings of over 30 seconds, reduces pipe blockage rate by over 90%, controls pressure and flow fluctuations within ±4%, and enables fully unmanned adaptive control. It solves long-standing technical problems in pumping high-strength concrete in ultra-high-rise buildings, significantly improving construction safety, continuity, and intelligence. It is applicable to high-strength concrete pumping construction in ultra-high-rise buildings over 200m in height.

Description

Technical Field

[0001] This invention relates to pumping of high-strength concrete in ultra-high-rise buildings, and in particular to a method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings. Background Technology

[0002] High-rise buildings generally use C60 and above high-strength concrete, which has high viscosity, high aggregate hardness, and a large amount of cementitious materials. Under vertical long-distance pumping conditions of more than 200m, it exhibits typical characteristics such as high frictional resistance along the pipeline, violent pressure fluctuations, easy abrupt changes in flow pattern, easy pipe blockage, and easy segregation. During the pumping process, pipeline pressure, pipe wall temperature, and concrete workability change in real time with height, time, and ambient temperature, belonging to a highly nonlinear, strongly coupled, time-varying complex system.

[0003] Existing pumping control technology has the following main drawbacks: 1) Relying solely on monitoring single-point parameters such as outlet pressure and flow rate cannot fully characterize the actual pumping flow state, resulting in delayed state identification and a high misjudgment rate; 2) There is a lack of a dedicated feature system and identification model for high-strength concrete in ultra-high-rise buildings, making it difficult to predict high-risk conditions such as pipe blockage and segregation in advance; 3) The control method mainly relies on manual experience or simple PID control, without an advance compensation mechanism, and the response to disturbances caused by height increase, viscosity change and temperature fluctuation is slow. 4) The lack of a multi-level collaborative mechanism involving the main pump, intermediate pressurization, and admixture delivery makes it impossible to achieve stable closed-loop control throughout the entire process.

[0004] The aforementioned problems have led to persistent challenges in the pumping construction of high-strength concrete in ultra-high-rise buildings, including high pipe blockage rates, significant segregation risks, delayed control, and poor continuity. These remain long-standing technical difficulties in this field that have yet to be effectively resolved. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying and adaptively controlling the pumping status of ultra-high-rise high-strength concrete.

[0006] The objective of this invention is achieved through the following technical solution: a method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings, comprising the following steps: S1: Install multi-source sensing units along the pipeline to collect multi-dimensional time-series sensing data, including friction pressure, pipe wall temperature, flow rate, vibration acceleration, and real-time viscosity. S2: Preprocess and extract features from the time-series sensing data to obtain a high-intensity pumping-specific feature set composed of pressure pulsation gradient, friction time-varying rate, temperature rise rate, vibration entropy, and viscosity-pressure correlation features. S3: Input the exclusive feature set into the hybrid model of time-varying friction mechanism and LSTM for state recognition, and output four pumping states: normal, abnormal friction increase, segregation precursor, and pipe blockage precursor, as well as early warning information. S4: Based on the identification results, perform multi-level collaborative adaptive regulation to adjust the main pump pressure and discharge, intermediate booster compensation pressure, and admixture micro-push amount in real time to achieve stable control of the pumping flow.

[0007] Optionally, the 3σ criterion is used to remove outliers in the time-series sensing data; cubic spline interpolation is used to complete missing data; and the data is mapped to the [0,1] interval by maximum-minimum normalization to eliminate dimensional differences.

[0008] Optionally, in step S2, feature extraction includes the following steps: S21: Extract time-domain features: calculate pressure pulsation gradient, flow rate fluctuation, temperature rise rate, vibration kurtosis, and viscosity transition coefficient; where, pressure pulsation gradient: ∇P(t) = (P(t) − P(t−Δt)) / Δt; Traffic volatility: Q_var(t) = |Q(t) − Q_avg| / Q_avg; Temperature rise rate: ΔT_rate(t) = (T(t) − T(t−Δt)) / Δt; Kurtosis: K(t) = E[x(t)] 4 ] / (E[x(t)²])²; Viscosity transition coefficient: Cμ(t) = |μ(t) − μ_ref| / μ_ref; Where: P(t) is the current pressure, Δt is the sampling step size, Q(t) is the current flow rate, Q_avg is the window average flow rate, T(t) is the current pipe wall temperature, x(t) is the vibration acceleration, μ(t) is the current viscosity, μ_ref is the reference viscosity, and E[・] is the mathematical expectation of the sliding window; S22: Constructing time-varying friction characteristics: Calculate the time-varying friction index R_fric(t) = ω1•dP(t) / dt + ω2•μ(t)•H(t), where μ(t) is the real-time viscosity, H(t) is the pumping height, dP(t) / dt is the pressure change rate, and ω1 and ω2 are the weighting coefficients specific to high-strength concrete; S23: Extracting frequency domain entropy features: Perform wavelet decomposition on the pressure and vibration sequences to obtain low-frequency trend components and high-frequency pulsation components, calculate the energy entropy of each component, and obtain the pumping chaotic features.

[0009] Optionally, in step S3, the hybrid model integrating the time-varying friction mechanism and LSTM is executed according to the following algorithm: S31: Time-varying friction mechanism branch: Instantaneous friction is calculated based on real-time viscosity μ(t), pressure change rate dP / dt, pumping speed v(t), pipe wall temperature difference ΔT(t), and pumping height H(t). τ(t) = k1•μ(t)+k2•dP / dt+k3•v(t)+k4•ΔT(t)+k5•H(t), and output the mechanism risk value R_mech(t); S32: LSTM Temporal Learning Branch: Input the time sequence of the dedicated feature set into the LSTM network, extract long temporal dependent features, and output the data-driven risk value R_lstm(t); S33: Dynamic weighted feature fusion: Calculate the fusion risk value R_fuse(t) = ω(t)•R_mech(t)+(1−ω(t))•R_lstm(t), where ω(t) is the dynamic weight; S34: Input the fusion risk value into the classifier and output the pumping status and early warning results.

[0010] The dynamic weight ω(t) is calculated using the following formula: ω(t) = σ( a•|τ(t)−τ_avg| + b•|dP / dt| ) In the formula: σ is the sigmoid function, τ_avg is the window average friction, and a and b are adaptive coefficients.

[0011] The hybrid model can provide an early warning of pipe blockage / segregation for at least 30 seconds and output the warning level and recommended control amount.

[0012] Optionally, in step S4, the multi-level collaborative adaptive control adopts model predictive control (MPC) and feedforward compensation to correct the control parameters in real time based on the pumping height, concrete characteristics, and ambient temperature.

[0013] Optionally, in step S4, the implementation of model predictive control (MPC) and feedforward compensation specifically includes: S41: Establish a time-varying prediction model for pumping high-strength concrete in super high-rise buildings. Using pumping height H (t), concrete viscosity μ (t), ambient temperature T_env (t), and frictional resistance τ (t) as inputs, predict the pumping pressure, flow rate, and flow regime deviation in the next N steps. S42: Construct feedforward compensation terms: Calculate the compensation amount in advance based on the static pressure increase caused by the increase in pumping height, the friction increase caused by viscosity change, and the performance degradation caused by temperature change, and add it to the control output in real time. S43: The MPC controller aims to minimize pressure fluctuations, ensure stable flow, prevent segregation, and avoid pipe blockage. It uses a rolling solution to obtain the optimal control law and outputs the main pump pressure / displacement, relay booster compensation pressure, and admixture delivery amount. S44: The feedforward compensation output and the MPC feedback output are superimposed to form the final multi-level collaborative control quantity, which corrects the pumping system parameters in real time.

[0014] The present invention has the following advantages: 1. This invention combines multi-dimensional information such as pressure, temperature, flow rate, vibration, and viscosity to accurately characterize flow regime changes with high accuracy and low hysteresis.

[0015] 2. By adopting a hybrid model of time-varying friction mechanism and LSTM, we can balance physical interpretability and temporal feature learning ability to achieve early warning of abnormal states and transform from passive processing to active prevention.

[0016] 3. Employing MPC + feedforward compensation multi-level coordinated control, it provides advance compensation for disturbances in height, viscosity, and temperature, resulting in high control accuracy and fast response, significantly reducing the risks of pipe blockage, segregation, and pipe burst.

[0017] 4. It realizes integrated closed-loop control of main pump, intermediate booster and admixture delivery, and is fully adaptive without manual intervention, improving pumping continuity, construction safety and efficiency.

[0018] 5. This invention is specifically adapted to high-intensity pumping scenarios in ultra-high-rise buildings of C60 and above and 200m and above, solving a long-standing technical problem in this field, and has significant technological progress and engineering application value. Detailed Implementation

[0019] A method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings, comprising the following steps: S1: Install multi-source sensing units along the pipeline to collect multi-dimensional time-series sensing data, including friction pressure, pipe wall temperature, flow rate, vibration acceleration, and real-time viscosity. S2: Preprocess and extract features from the time-series sensing data to obtain a high-intensity pumping-specific feature set composed of pressure pulsation gradient, friction time-varying rate, temperature rise rate, vibration entropy, and viscosity-pressure correlation features. S3: Input the exclusive feature set into the hybrid model of time-varying friction mechanism and LSTM for state recognition, and output four pumping states: normal, abnormal friction increase, segregation precursor, and pipe blockage precursor, as well as early warning information. S4: Based on the identification results, perform multi-level collaborative adaptive regulation to adjust the main pump pressure and discharge, intermediate booster compensation pressure, and admixture micro-push amount in real time to achieve stable control of the pumping flow.

[0020] In this embodiment, the 3σ criterion is used to remove outliers in the time-series sensing data; cubic spline interpolation is used to complete the missing data; and the data is mapped to the [0,1] interval by maximum-minimum normalization to eliminate dimensional differences.

[0021] In this embodiment, feature extraction in step S2 includes the following steps: S21: Extract time-domain features: calculate pressure pulsation gradient, flow rate fluctuation, temperature rise rate, vibration kurtosis, and viscosity transition coefficient; where, pressure pulsation gradient: ∇P(t) = (P(t) − P(t−Δt)) / Δt; Traffic volatility: Q_var(t) = |Q(t) − Q_avg| / Q_avg; Temperature rise rate: ΔT_rate(t) = (T(t) − T(t−Δt)) / Δt; Kurtosis: K(t) = E[x(t)] 4 ] / (E[x(t)²])²; Viscosity transition coefficient: Cμ(t) = |μ(t) − μ_ref| / μ_ref; Where: P(t) is the current pressure, Δt is the sampling step size, Q(t) is the current flow rate, Q_avg is the window average flow rate, T(t) is the current pipe wall temperature, x(t) is the vibration acceleration, μ(t) is the current viscosity, μ_ref is the reference viscosity, and E[・] is the mathematical expectation of the sliding window; S22: Constructing time-varying friction characteristics: Calculate the time-varying friction index R_fric(t) = ω1•dP(t) / dt + ω2•μ(t)•H(t), where μ(t) is the real-time viscosity, H(t) is the pumping height, dP(t) / dt is the pressure change rate, and ω1 and ω2 are the weighting coefficients specific to high-strength concrete; S23: Extracting frequency domain entropy features: Perform wavelet decomposition on the pressure and vibration sequences to obtain low-frequency trend components and high-frequency pulsation components, calculate the energy entropy of each component, and obtain the pumping chaotic features.

[0022] In this embodiment, in step S3, the hybrid model integrating the time-varying friction mechanism and LSTM is executed according to the following algorithm: S31: Time-varying friction mechanism branch: Instantaneous friction is calculated based on real-time viscosity μ(t), pressure change rate dP / dt, pumping speed v(t), pipe wall temperature difference ΔT(t), and pumping height H(t). τ(t) = k1•μ(t)+k2•dP / dt+k3•v(t)+k4•ΔT(t)+k5•H(t), and output the mechanism risk value R_mech(t); S32: LSTM Temporal Learning Branch: Input the time sequence of the dedicated feature set into the LSTM network, extract long temporal dependent features, and output the data-driven risk value R_lstm(t); S33: Dynamic weighted feature fusion: Calculate the fusion risk value R_fuse(t) = ω(t)•R_mech(t)+(1−ω(t))•R_lstm(t), where ω(t) is the dynamic weight; S34: Input the fusion risk value into the classifier and output the pumping status and early warning results.

[0023] The dynamic weight ω(t) is calculated using the following formula: ω(t) = σ( a•|τ(t)−τ_avg| + b•|dP / dt| ) In the formula: σ is the sigmoid function, τ_avg is the window average friction, and a and b are adaptive coefficients.

[0024] The hybrid model can provide an early warning of pipe blockage / segregation for at least 30 seconds and output the warning level and recommended control amount.

[0025] In this embodiment, in step S4, the multi-level collaborative adaptive control adopts model predictive control (MPC) and feedforward compensation to correct the control parameters in real time based on the pumping height, concrete characteristics, and ambient temperature.

[0026] In this embodiment, step S4, specifically including the implementation of model predictive control (MPC) and feedforward compensation, includes: S41: Establish a time-varying prediction model for pumping high-strength concrete in super high-rise buildings. Using pumping height H (t), concrete viscosity μ (t), ambient temperature T_env (t), and frictional resistance τ (t) as inputs, predict the pumping pressure, flow rate, and flow regime deviation in the next N steps. S42: Construct feedforward compensation terms: Calculate the compensation amount in advance based on the static pressure increase caused by the increase in pumping height, the friction increase caused by viscosity change, and the performance degradation caused by temperature change, and add it to the control output in real time. S43: The MPC controller aims to minimize pressure fluctuations, ensure stable flow, prevent segregation, and avoid pipe blockage. It uses a rolling solution to obtain the optimal control law and outputs the main pump pressure / displacement, relay booster compensation pressure, and admixture delivery amount. S44: The feedforward compensation output and the MPC feedback output are superimposed to form the final multi-level collaborative control quantity, which corrects the pumping system parameters in real time.

[0027] Taking a super high-rise office building project in Southwest China as an example, the building is a steel structure-concrete core tube hybrid structure. The core tube walls are made of C60 high-strength concrete. The design pumping height is 0m-428m for continuous vertical pumping, with a single batch pumping volume of 180m³-260m³. The pipeline uses Φ150mm high-pressure wear-resistant pump pipes, laid along the inner wall of the core tube. Two intermediate booster stations are set up, located at heights of 150m and 300m respectively. The construction of this project has the following difficulties: the concrete is C60... High-strength concrete with a water-cement ratio of 0.31 has high viscosity and friction, resulting in a >35% probability of pipe blockage during conventional pumping. The pumping height is 428m, with a static head pressure as high as approximately 9.8MPa, leading to significant pressure loss and drastic pressure fluctuations along the pumping route. Construction takes place in summer with ambient temperatures ranging from 24℃ to 38℃, resulting in large diurnal temperature variations and rapid degradation of concrete workability. The construction of ultra-high-rise buildings requires continuous operation without prolonged downtime, placing extremely high demands on system stability and fault tolerance. Traditional manual pressure regulation methods frequently result in pressure shocks, unstable flow rates, pipe blockages, and segregation, with each blockage treatment taking 2-5 hours, severely impacting the construction period and structural quality.

[0028] The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings, as described in this invention, enables unmanned, adaptive, and stable pumping throughout the entire process, fully verifying the technical advantages and engineering applicability of this invention.

[0029] According to the technical solution of this invention, a complete sensing, identification, and control system is deployed at the construction site: Sensing unit deployment: Friction pressure sensor: 1 channel is installed every 50m, for a total of 9 channels, with a range of 0-40MPa and an accuracy of ±0.1% FS; Pipe wall temperature sensor: installed at the same location as the pressure sensor, with a total of 9 channels, measuring range -20℃ to 120℃, and accuracy ±0.1℃; Pump outlet flow sensor: 1 channel, electromagnetic flow meter, range 0-180m³ / h; Vibration acceleration sensor: 1 channel each at pump outlet, 150m relay station, and 300m relay station, with a range of ±20g and a sampling frequency of 1kHz; Microwave online viscosity sensor: 1 channel at pump outlet, real-time measurement of apparent viscosity of concrete, range 0-1000 Pa·s; Height acquisition module: Calculates pumping height H(t) in real time based on pump stroke counter; Ambient temperature sensor: 1 channel at the construction site, collecting ambient temperature T_env(t).

[0030] Data Acquisition and Control Platform: Industrial data acquisition cabinet: supports 16 analog inputs, 8 control outputs, and a sampling frequency of 10Hz; Edge computing unit: Equipped with a GPU, running a hybrid recognition model and MPC controller with a latency of <100ms; Actuator interface: Enables Modbus TCP communication with the main pump PLC, relay booster station, and online admixture delivery system.

[0031] Implementing agency: Main pump: High-pressure plunger pump, rated pressure 35MPa, adjustable displacement; Intermediate booster stations: 2 sets, pressure compensation range 0-16MPa; Online admixture delivery system: High-precision metering pump, delivery range 0-12L / min, with micro-dose continuous adjustment.

[0032] The detailed steps are as follows: Step S1: Multi-source time-series data acquisition. After system startup, synchronous sampling at 10Hz is used to collect the following data in real time and cache the time-series sequence with a sliding window length of 300 points (30s): P1(t)—P9(t): Pipeline pressure at 0m, 50m, 100m, 150m, 200m, 250m, 300m, 350m, and 400m; T1(t)—T9(t): Pipeline wall temperature at the corresponding point; Q(t): Instantaneous flow rate at pump outlet; x1(t), x2(t), x3(t): Vibration acceleration at pump outlet, 150m, and 300m; μ(t): Real-time concrete viscosity; H(t): Real-time pumping height; T_env(t): Ambient temperature. All data are timestamped to form a multi-dimensional, long-time-series, and spatiotemporally correlated pumping dataset, providing complete input for subsequent preprocessing, feature extraction, and state recognition.

[0033] Step S2: Data Preprocessing 1) Outlier removal: using the 3σ criterion Calculate the mean μ_win and standard deviation σ_win for the data within the sliding window. If the data at a certain moment satisfies: |x (t) − μ_win| > 3σ_win If it is determined to be an outlier (caused by vibration, electromagnetic interference, or instantaneous impact), the window mean will be used to replace it.

[0034] In this embodiment, the pump reversal impact occasionally causes instantaneous pressure jumps. After being eliminated by the 3σ criterion, the smoothness of the data is significantly improved.

[0035] 2) Missing value interpolation: Cubic spline interpolation is used. To address missing data caused by instantaneous packet loss in network transmission and sensors, cubic spline interpolation is used to complete the data, ensuring the continuity of the time series without any breaks. In this embodiment, the measured packet loss rate is <0.3%, and there is no significant distortion after interpolation.

[0036] 3) Maximum-Minimum Normalization Linearly map all data to the interval [0, 1] to eliminate dimensional differences in pressure (MPa), temperature (°C), flow rate (m³ / h), vibration (g), and viscosity (Pa·s). The formula is: x'_norm = (x_raw − x_min) / (x_max − x_min) Where x_min and x_max are the extreme values ​​of each sensor's range.

[0037] After preprocessing, the data has no outliers, no missing values, and no dimensional differences, fully meeting the model input requirements.

[0038] Step S3: High-intensity pumping-specific feature extraction: Temporal feature extraction: 1) Pressure pulsation gradient ∇P(t); Formula: ∇P(t) = (P(t) − P(t−Δt)) / Δt, In this embodiment, Δt=0.1s, and the pump outlet pressure P1(t) is used for calculation. It is used to characterize the intensity of instantaneous pump pressure fluctuation, directly reflecting the precursors of pipe blockage and sudden changes in flow. Under normal operating conditions: ∇P(t) is stable within ±0.2MPa / s; Friction increases: ∇P(t) rises to 0.5—1.2MPa / s; Precursors of pipe blockage: ∇P(t) > 1.5MPa / s and continues to rise.

[0039] 2) Flow fluctuation rate Q_var(t); Formula: Q_var(t) = |Q(t) − Q_avg| / Q_avg, where Q_avg is the average flow rate over a 30-second window, used to quantify pumping stability. Normal: Q_var < 5%; Abnormal: Q_var > 12%, indicating uneven pump discharge or localized blockage in the pipeline.

[0040] 3) Temperature rise rate ΔT_rate(t), formula: ΔT_rate(t) = (T(t) − T(t−Δt)) / Δt, Δt=0.1s, calculated using temperature T7(t) at a height of 300m. It reflects the intensity of frictional heat generation and is an important indicator of abnormal friction resistance. Normal: <0.05℃ / s; High friction resistance: >0.1℃ / s, indicating that the pipe is about to enter the critical state of blockage.

[0041] 4) Kujicic K(t), formula: K(t) = E[x(t)]4 ] / (E [x (t)²])², where x (t) is the pump outlet vibration acceleration, and E [・] is the mathematical expectation of the 30s window. Kurtosis is extremely sensitive to impact, jamming, and particle accumulation. Normal: 3-5; Precursor to jamming / blockage: >8, with sharp impact characteristics.

[0042] 5) Viscosity transition coefficient Cμ(t) Formula: Cμ(t) = |μ(t) − μ_ref| / μ_ref, where μ_ref is the factory reference viscosity, used to judge the decline in workability and the trend of segregation. Normal: <10%; Precursor to segregation: >25%, indicating that an additive is needed for compensation.

[0043] Time-varying friction characteristics: The formula for calculating the time-varying friction index R_fric(t) is: R_fric(t) = ω1・dP(t) / dt + ω2・μ(t)・H(t). In this embodiment, ω1=0.4 and ω2=0.6 are calibrated, specifically for C60 concrete. This feature integrates pressure change rate, real-time viscosity, and pumping height, and is a unique core feature of high-strength pumping in ultra-high-rise buildings. It can directly characterize the overall friction level of the pipeline, solving the defect of traditional methods that cannot judge the friction of the entire pipeline by relying solely on single-point pressure. Normal pumping: R_fric is stable within the threshold range; abnormal increase in friction: R_fric continues to rise, and the slope increases; critical blockage: R_fric exceeds the upper limit and is accompanied by oscillation.

[0044] Frequency domain entropy characteristics: Wavelet decomposition was performed on pressure P1(t) and vibration x1(t) to obtain low-frequency trend components and high-frequency pulsation components. The energy entropy of each component was calculated to reflect the degree of chaos in the pumping flow: low energy entropy indicates a stable and orderly flow; high energy entropy indicates a turbulent flow, uneven friction, and a significant tendency for segregation. Finally, five time-domain features, one time-varying friction feature, and six frequency-domain entropy features were concatenated to form a 12-dimensional high-intensity pumping-specific feature set, which was used as input to the hybrid model.

[0045] In this implementation, the time-varying friction mechanism is integrated with the LSTM hybrid model for state recognition. By using a dual-branch dynamic weighted hybrid model, the interpretability of the physical mechanism and the long-term learning capability of LSTM are taken into account, thus solving the industry pain points of insufficient accuracy of pure mechanism models and uninterpretability of pure data models.

[0046] 1) Time-varying friction mechanism branch: Formula: τ(t) = k1・μ(t)+k2・dP / dt+k3・v(t)+k4・ΔT(t)+k5・H(t) This embodiment is based on the C60 concrete calibration coefficients: k1=0.28, k2=0.22, k3=0.15, k4=0.10, k5=0.25. The inputs are: real-time viscosity, pressure change rate, pumping speed, pipe wall temperature difference, and pumping height. The output is: mechanism risk value R_mech(t), in the range [0, 1]. The higher the value, the greater the risk.

[0047] 2) LSTM temporal learning branch, input: 300-point × 12-dimensional feature temporal sequence, network structure: input layer → normalization layer → 2-layer LSTM (64 neurons + 32 neurons) → Dropout layer → fully connected layer, output: data-driven risk value R_lstm(t), range [0, 1]. LSTM can learn long-term temporal dependencies, capture early weak anomaly features, and achieve early warning.

[0048] 3) Dynamically weighted feature fusion: Formula: R_fuse(t) = ω(t)・R_mech(t) + (1−ω(t))・R_lstm(t) Dynamic weights: ω(t) = σ( a・|τ(t)−τ_avg| + b・|dP / dt| ), where σ is the sigmoid function, a=1.2, b=0.8, and τ_avg is the window average friction.

[0049] Stable operating conditions: Small frictional fluctuations → Reduced ω(t) → Increased LSTM weights, resulting in higher recognition accuracy; Dramatic changes in operating conditions: Frictional resistance increases sharply → ω(t) increases → Mechanism weight increases, ensuring physical consistency and safety.

[0050] 4) Classification and Early Warning Output Input R_fuse(t) into the Softmax classifier, and it will output four states: ① Normal; ② Abnormally high friction; ③ Precursor to segregation; ④ Precursor to pipe blockage.

[0051] Simultaneously output: warning level, confidence level, and advance warning time.

[0052] The measured advance warning time in this embodiment is 32s-45s, which meets the requirements of active control.

[0053] 6.2 Typical Recognition Results of This Embodiment 1) Pumping height 180m, ambient temperature 32℃, concrete viscosity slightly increased, model output: abnormal increase in friction, warning level I, confidence level 94%, 36s ahead, the system automatically started small pressure compensation, and did not develop into pipe blockage.

[0054] 2) Pumping height 320m, long-term high friction caused the viscosity to rise rapidly. Model output: pre-segregation sign, warning level II, confidence level 96%, 32s ahead. The system immediately started micro-dose delivery of admixture to stabilize operation.

[0055] 3) Extreme working condition: slight localized aggregate blockage. Model output: early warning of blockage, warning level III, confidence level 98%, 41 seconds ahead. The system quickly reduces the discharge, increases the relay pressure, and performs small-scale pulsation flushing to successfully remove the blockage.

[0056] In step S4, multi-level coordinated adaptive regulation is implemented. In this embodiment, a composite architecture of model predictive control (MPC) and feedforward compensation is adopted to achieve three-level coordinated adaptive regulation of the main pump, intermediate booster, and admixture. 1) Height feedforward compensation ΔP1(t), formula: ΔP1(t) = ρ・g・H (t), ρ=2450kg / m³ (C60 concrete density), g=9.8m / s², H (t) is the real-time height. Function: to compensate for the static head pressure caused by the increase in height, and to avoid the flow rate decrease due to insufficient pressure. At a height of 428m, ΔP1≈10.3MPa.

[0057] 2) Viscosity feedforward compensation ΔP2(t), formula: ΔP2(t) = kμ・μ(t)・L, where L is the total length of the pipeline and kμ is the friction-viscosity coefficient, calibrated according to C60. Function: to compensate for friction loss along the pipeline caused by high viscosity and ensure flow stability.

[0058] 3) Temperature feedforward compensation ΔP3(t), formula: ΔP3(t) = kT・(T_env (t) − T0), T0=25℃ (reference temperature), kT is the temperature compensation coefficient, function: to compensate for the performance degradation caused by high temperature and reduce the risk of segregation.

[0059] Total feedforward output: U_ff(t) = ΔP1(t) + ΔP2(t) + ΔP3(t) Model Predictive Control (MPC): 1) Prediction model: Based on the current state: P(t), Q(t), τ(t), μ(t), predict the changes in pressure, flow rate, and friction in the next **N=20 steps (2s)**.

[0060] 2) Optimize the objective function, min J = w1・Σ(P_pred−P_ref)² + w2・Σ(Q_pred−Q_ref)² + w3・ΣΔu² Weights: w1=0.5, w2=0.3, w3=0.2; Objective: Stable pressure, stable flow, and no impact on control volume.

[0061] 3) Constraints: Main pump pressure: ≤32MPa; Intermediate pump pressure: ≤14MPa; Admixture delivery rate: 0-10L / min; Flow rate: 40-120 m³ / h.

[0062] 4) MPC output: U_fb(t) (feedback control quantity) Total control volume and allocation of three-level actuators Total control quantity: U_total(t) = U_ff(t) + U_fb(t) Allocation logic: 1) Main pump: Adjusts the discharge capacity and outlet pressure to ensure stable total flow; 2) Relay booster stations: Compensation for pressure loss along the pipeline is segmented by height, with 30% compensation for sections below 150m, 40% for sections between 150 and 300m, and 30% for sections above 300m. 3) Additive system: Micro-dose delivery is initiated only when there are signs of segregation, with a delivery rate of 0.5-3 L / min, to maintain viscosity stability.

[0063] In this embodiment, the pumping height increases from 0m to 428m. The system automatically feeds forward to compensate for static pressure, and the flow fluctuation is always controlled within ±3%, which is far superior to the traditional ±15%-30%. When the viscosity of C60 concrete fluctuates, the system adjusts the pressure and admixtures in real time, and the viscosity transition coefficient is controlled within 10%, with no segregation or bleeding. When the ambient temperature rises from 24℃ to 38℃, the system automatically increases temperature compensation and admixture delivery, maintaining stable performance without manual adjustment. The main pump and the two-stage relay stations are smoothly linked, with no pressure shock, no overshoot, and no interruption. Continuous pumping can be carried out for 12 hours without downtime.

[0064] This embodiment compares the method of the present invention with traditional manual voltage regulation and conventional PID control under the same operating conditions, and the results are as follows: index Traditional manual control Conventional PID control Method of the present invention Increase Pipe blockage rate 32.7% 18.3% 1.8% Reduced by 94.5% segregation incidence 15.2% 9.7% 0.9% Reduced by 94.1% Pressure fluctuations ±18%—±28% ±12%—±20% ±2%—±4% Significantly stable Traffic fluctuations ±15%—±30% ±10%—±18% ±2%—±4% Significantly stable Advanced warning time 0s (afterwards) 0—5s 32—45s Advanced warning Number of manual interventions 1 time / 30 minutes 1 time / 60min 0 times / Full journey Unmanned Continuous pumping time 2.5h 5h ＞12h Significantly improved Construction efficiency Benchmark 100% 115% 187% Increased by 87% Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings, characterized in that: Includes the following steps: S1: Install multi-source sensing units along the pipeline to collect multi-dimensional time-series sensing data, including friction pressure, pipe wall temperature, flow rate, vibration acceleration, and real-time viscosity. S2: Preprocess and extract features from the time-series sensing data to obtain a high-intensity pumping-specific feature set composed of pressure pulsation gradient, friction time-varying rate, temperature rise rate, vibration entropy, and viscosity-pressure correlation features. S3: Input the exclusive feature set into the hybrid model of time-varying friction mechanism and LSTM for state recognition, and output four pumping states: normal, abnormal friction increase, segregation precursor, and pipe blockage precursor, as well as early warning information. S4: Based on the identification results, perform multi-level collaborative adaptive regulation to adjust the main pump pressure and discharge, intermediate booster compensation pressure, and admixture micro-push amount in real time to achieve stable control of the pumping flow.

2. The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings according to claim 1, characterized in that: The 3σ criterion was used to remove outliers from the time-series sensing data; cubic spline interpolation was used to complete the missing data; and the data was mapped to the [0,1] interval by maximum-minimum normalization to eliminate dimensional differences.

3. The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings according to claim 2, characterized in that: In step S2, feature extraction includes the following steps: S21: Extract time-domain features: calculate pressure pulsation gradient, flow rate fluctuation, temperature rise rate, vibration kurtosis, and viscosity transition coefficient; where, pressure pulsation gradient: ∇P(t) = (P(t) − P(t−Δt)) / Δt; Traffic volatility: Q_var(t) = |Q(t) − Q_avg| / Q_avg; Temperature rise rate: ΔT_rate(t) = (T(t) − T(t−Δt)) / Δt; Kurtosis: K(t) = E[x(t)] 4 ] / (E[x(t)²])²; Viscosity transition coefficient: Cμ(t) = |μ(t) − μ_ref| / μ_ref; Where: P(t) is the current pressure, Δt is the sampling step size, Q(t) is the current flow rate, Q_avg is the window average flow rate, T(t) is the current pipe wall temperature, x(t) is the vibration acceleration, μ(t) is the current viscosity, μ_ref is the reference viscosity, and E[・] is the mathematical expectation of the sliding window; S22: Constructing time-varying friction characteristics: Calculate the time-varying friction index R_fric(t) = ω1·dP(t) / dt + ω2·μ(t)·H(t), where μ(t) is the real-time viscosity, H(t) is the pumping height, dP(t) / dt is the pressure change rate, and ω1 and ω2 are the weighting coefficients specific to high-strength concrete; S23: Extracting frequency domain entropy features: Perform wavelet decomposition on the pressure and vibration sequences to obtain low-frequency trend components and high-frequency pulsation components, calculate the energy entropy of each component, and obtain the pumping chaotic features.

4. The method for identifying and adaptively controlling the pumping status of ultra-high-rise high-strength concrete according to claim 3, characterized in that: In step S3, the hybrid model integrating the time-varying friction mechanism and LSTM is executed according to the following algorithm: S31: Time-varying friction mechanism branch: Instantaneous friction is calculated based on real-time viscosity μ(t), pressure change rate dP / dt, pumping speed v(t), pipe wall temperature difference ΔT(t), and pumping height H(t). τ(t) = k1·μ(t)+k2·dP / dt+k3·v(t)+k4·ΔT(t)+k5·H(t), and output the mechanism risk value R_mech(t); S32: LSTM Temporal Learning Branch: Input the time sequence of the dedicated feature set into the LSTM network, extract long temporal dependent features, and output the data-driven risk value R_lstm(t); S33: Dynamic weighted feature fusion: Calculate the fusion risk value R_fuse(t) = ω(t)·R_mech(t)+(1−ω(t))·R_lstm(t), where ω(t) is the dynamic weight; S34: Input the fusion risk value into the classifier and output the pumping status and early warning results.

5. The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings according to claim 4, characterized in that: The dynamic weight ω(t) is calculated using the following formula: ω(t) = σ( a·|τ(t)−τ_avg| + b·|dP / dt| ) In the formula: σ is the sigmoid function, τ_avg is the window average friction, and a and b are adaptive coefficients.

6. The method for identifying and adaptively controlling the pumping status of ultra-high-rise high-strength concrete according to claim 5, characterized in that: The hybrid model can provide an early warning of pipe blockage / segregation for at least 30 seconds and output the warning level and recommended control amount.

7. The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings according to claim 6, characterized in that: In step S4, the multi-level collaborative adaptive control adopts model predictive control (MPC) and feedforward compensation to correct the control parameters in real time based on the pumping height, concrete characteristics, and ambient temperature.

8. The method for identifying and adaptively controlling the pumping status of high-strength concrete in ultra-high-rise buildings according to claim 7, characterized in that: In step S4, the implementation of model predictive control (MPC) and feedforward compensation specifically includes: S41: Establish a time-varying prediction model for pumping high-strength concrete in super high-rise buildings. Using pumping height H (t), concrete viscosity μ (t), ambient temperature T_env (t), and frictional resistance τ (t) as inputs, predict the pumping pressure, flow rate, and flow regime deviation in the next N steps. S42: Construct feedforward compensation terms: Calculate the compensation amount in advance based on the static pressure increase caused by the increase in pumping height, the friction increase caused by viscosity change, and the performance degradation caused by temperature change, and add it to the control output in real time. S43: The MPC controller aims to minimize pressure fluctuations, ensure stable flow, prevent segregation, and avoid pipe blockage. It uses a rolling solution to obtain the optimal control law and outputs the main pump pressure / displacement, relay booster compensation pressure, and admixture delivery amount. S44: The feedforward compensation output and the MPC feedback output are superimposed to form the final multi-level coordinated control quantity, which corrects the pumping system parameters in real time.