A Method and System for Treating Food Wastewater Based on Multidimensional Data Diagnosis

Abstract

This invention discloses a method and system for treating kitchen waste wastewater based on multidimensional data diagnosis, belonging to the field of wastewater treatment technology. The method includes: acquiring the feed parameters of the kitchen waste and the real-time operating parameters of the anaerobic reactor, and calculating the oil load characteristics of the kitchen waste wastewater. Based on the oil load characteristics, a variable-speed shear scan is performed on the stirring device of the anaerobic reactor, and the response sequence of stirring torque changing with rotational speed is collected. The oil encapsulation level is calculated based on the response sequence. Multi-component concentration data of volatile fatty acids in the effluent of the anaerobic reactor are collected, and the multi-component concentration data is constructed into a multidimensional vector sequence. The status diagnosis results of the acid-producing bacteria community function are obtained based on the multidimensional vector sequence. Based on the oil encapsulation level and status diagnosis results, adjustment instructions for the stirring strategy and control instructions for the feed are generated. This invention can ensure the treatment efficiency of kitchen waste wastewater.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment technology, specifically relating to a method and system for treating kitchen waste wastewater based on multidimensional data diagnosis. Background Technology

[0002] The mainstream treatment process for food waste is "pretreatment—anaerobic digestion—aerobic biological treatment—advanced treatment—discharge meeting standards". In the anaerobic digestion stage, organic wastewater enters the anaerobic reactor. Under the synergistic action of hydrolytic acidifying bacteria and methanogenic bacteria, macromolecular organic matter is gradually degraded into volatile fatty acids and further converted into methane and carbon dioxide, achieving a significant reduction in organic matter and the recovery and utilization of biogas energy. However, the above conventional process suffers from unstable anaerobic digestion efficiency in actual operation. To address these issues, existing technologies utilize online sensors to monitor macroscopic parameters such as pH, redox potential, gas production rate, and total volatile fatty acid concentration within the anaerobic reactor in real time. Based on the trends in these parameters, the feed rate and stirring intensity are adjusted to ensure the stability of the anaerobic digestion process.

[0003] However, food waste contains a large amount of animal and vegetable oils. These oils do not exist in a uniformly dispersed form within the anaerobic reactor. Instead, they form oil droplets or films in areas with insufficient stirring and shear force, gradually coating the surface of solid organic particles and forming a hydrophobic physical barrier. This oil coating phenomenon significantly reduces the accessibility of the coated substrate to hydrolytic enzymes and acid-producing bacteria, resulting in areas within the reactor where easily degradable substrates are rapidly consumed, and areas where substrates are shielded and remain for extended periods. Under these conditions, the population structure of acid-producing bacteria gradually shifts towards rapidly proliferating strains that prefer simple carbohydrate metabolism. In the early to mid-stages of this shift, due to the rapid acidification of carbohydrate substrates, macroscopic monitoring indicators such as total volatile fatty acid production and gas production can still remain within normal ranges, and existing monitoring methods cannot detect the significant distortion of the bacterial community's functional structure in a timely manner. After the offset accumulates to a certain extent, a large amount of insufficiently acidified oil and protein enters the subsequent treatment unit with the anaerobic effluent. This not only leads to a persistently high COD in the aerobic system effluent, but the accumulated long-chain fatty acids also inhibit the activity of methanogens, ultimately causing the overall performance of the anaerobic digestion system to deteriorate. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a method and system for treating kitchen waste and wastewater based on multidimensional data diagnosis, thereby resolving the issues present in the background art.

[0005] To achieve the aforementioned objectives, this invention proposes a method for treating kitchen waste wastewater based on multidimensional data diagnosis, comprising: The feed parameters of food waste and the real-time operating parameters of the anaerobic reactor were obtained, and the oil load characteristics of the food waste wastewater were calculated based on the feed parameters and real-time operating parameters. Based on the characteristics of oil loading, a variable-speed shear scan was performed on the stirring device of the anaerobic reactor, and the response sequence of stirring torque as a function of rotational speed was collected. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate was calculated. Multi-component concentration data of volatile fatty acids in the effluent of an anaerobic reactor were collected. The multi-component concentration data within a continuous time window were constructed into a multi-dimensional vector sequence. Based on the trajectory morphology characteristics of the multi-dimensional vector sequence in the component space, the functional status diagnosis results of the acid-producing bacteria were obtained. Based on the oil coating level and state diagnosis results, adjustment instructions for the stirring strategy and control instructions for the feed are generated. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

[0006] This invention also provides a food waste wastewater treatment system based on multidimensional data diagnosis. This system is used to implement the methods described above, and includes: The operating condition sensing module acquires the feed parameters of kitchen waste and the real-time operating parameters of the anaerobic reactor, and calculates the oil load characteristics of kitchen waste wastewater based on the feed parameters and real-time operating parameters. The physical property detection module performs variable-speed shear scanning on the stirring device of the anaerobic reactor based on the characteristics of oil loading, and collects the response sequence of stirring torque as the speed changes. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate is calculated. The biochemical diagnostic module collects multi-component concentration data of volatile fatty acids in the effluent of the anaerobic reactor, constructs multi-dimensional vector sequences from the multi-component concentration data within a continuous time window, and obtains the functional status diagnostic results of acid-producing bacteria based on the trajectory morphology characteristics of the multi-dimensional vector sequences in the component space. The intelligent decision-making module generates adjustment instructions for the stirring strategy and control instructions for the feed based on the oil coating level and state diagnosis results. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

[0007] The beneficial effects of this invention are as follows: This invention first calculates the oil load characteristics by acquiring feed and operating parameters, and then performs variable-speed shear scanning on the stirring device to collect torque response sequences, thereby assessing the oil's encapsulation level on the solid substrate. Next, it constructs a multi-dimensional vector sequence by collecting volatile fatty acid concentrations and extracts its component trajectory morphology characteristics, overcoming the lag of traditional detection methods and achieving early and accurate diagnosis of the functional state of acid-producing bacteria. By integrating the encapsulation level and diagnostic results, it generates stirring and feed control commands and predicts the residual load after control, outputting adjustment suggestions downstream, thus achieving coordinated anti-impact control throughout the entire process. Notably, this invention solves the problem in existing technologies where the latent distortion of bacterial function caused by oil encapsulation of the substrate cannot be detected in time by conventional indicators, ultimately leading to the deterioration of the overall system performance. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the steps of a method for treating kitchen waste and wastewater based on multidimensional data diagnosis according to the present invention. Figure 2 This is a schematic diagram illustrating the principle of generating the residence density map in this invention; Figure 3 This is a schematic diagram illustrating the application effect of the present invention; Figure 4 This is a schematic diagram of the structure of the kitchen waste and wastewater treatment system based on multidimensional data diagnosis according to the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0010] It is understood that the terms "first," "second," etc., used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script.

[0011] like Figure 1 As shown, a method for treating kitchen waste wastewater based on multidimensional data diagnosis includes: S1: Obtain the feed parameters of food waste and the real-time operating parameters of the anaerobic reactor, and calculate the oil load characteristics of food waste wastewater based on the feed parameters and real-time operating parameters.

[0012] First, the equipment architecture is introduced. The food waste resource recovery and wastewater treatment system in this embodiment includes a pretreatment unit, an anaerobic digestion unit, an aerobic biological treatment unit, and an advanced treatment unit. The pretreatment unit sorts, crushes, and performs preliminary oil-water separation on the collected food waste, converting it into a homogeneous slurry suitable for subsequent biological treatment. The anaerobic digestion unit, under anaerobic conditions, uses microbial metabolism to gradually decompose the organic matter in the slurry into volatile fatty acids, ultimately converting it into biogas, achieving both organic matter reduction and energy recovery. The aerobic treatment unit further biodegrades and nitrifies the residual organic matter and ammonia nitrogen in the anaerobic digestion effluent, ensuring the effluent quality meets the influent requirements for subsequent advanced treatment. The advanced treatment unit performs final water quality refinement on the aerobic treatment effluent, removing residual suspended solids, color, and pathogenic microorganisms, ensuring the final effluent consistently meets discharge standards.

[0013] Feed parameters represent the source and composition characteristics of wastewater materials, while real-time operating parameters represent the internal environmental state and biochemical reaction stability characteristics of the anaerobic reactor. The oil load characteristics calculated based on these two parameters represent the dynamic accumulation level of oil and the physical shielding risk to substrate degradation. The calculation method for oil load characteristics will be described in detail later.

[0014] S2: Based on the characteristics of oil loading, a variable-speed shear scan is performed on the stirring device of the anaerobic reactor, and the response sequence of stirring torque as the speed changes is collected. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate is calculated.

[0015] This step, by actively altering the shear field and capturing torque feedback, enables the acquisition of viscoelastic changes in materials and their sensitivity to shear force, thus overcoming the hysteresis problem inherent in traditional chemical sensors. The response sequence records the torque feedback patterns of the material at different shear rates, used to describe the rheological properties of the mixture within the reactor. The grease encapsulation level represents the severity of physical barriers formed by grease on the surface of solid particles, reflecting the degree of contact obstruction between the substrate and microorganisms and enzymes.

[0016] S3: Collect multi-component concentration data of volatile fatty acids in the effluent of the anaerobic reactor, construct a multi-dimensional vector sequence from the multi-component concentration data within a continuous time window, and obtain the functional status diagnosis results of the acid-producing bacteria community based on the trajectory morphology characteristics of the multi-dimensional vector sequence in the component space.

[0017] The effluent from the anaerobic reactor was sampled and analyzed using gas chromatography to obtain the concentration values ​​of four volatile fatty acids: acetic acid, propionic acid, butyric acid, and valeric acid. Since the component data at a single sampling time only reflects the instantaneous state and is easily affected by random fluctuations, data from multiple sampling times were selected to construct a multidimensional vector sequence. The component space is a multidimensional mathematical space, and each vector in the multidimensional vector sequence corresponds to a coordinate point in the component space. Connecting the coordinate points sequentially in time forms the component trajectory. The spatial orientation and deformation of the component trajectory reflect the dynamic changes in the proportional relationships between the various fatty acid components over time. Finally, the functional status diagnosis results of the acid-producing bacteria were extracted based on the component trajectory. These results provide a basis for decision-making regarding subsequent adjustments to the stirring strategy and control of the feed group distribution ratio.

[0018] S4: Based on the oil coating level and state diagnosis results, generate adjustment instructions for the stirring strategy and control instructions for the feed. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

[0019] After obtaining the oil coating level and state diagnosis results, the severity of the physical barrier of oil to solid substrate is determined based on the oil coating level, and the metabolic health status of microorganisms is determined based on the state diagnosis results. Then, based on the two, adjustment instructions for stirring strategy and control instructions for feed are generated. By adjusting the stirring parameters, the oil coating layer in the reactor is broken, and by controlling the composition ratio of the feed, the newly input organic load is controlled, thereby ensuring the efficient operation of the anaerobic reactor.

[0020] In this embodiment, the oil load characteristics of kitchen waste wastewater are calculated based on feed parameters and real-time operating parameters, including: Feed parameters include weight, crude fat content, and source category label, while real-time operating parameters include temperature and liquid level data.

[0021] A near-infrared spectroscopy analyzer is installed in the sorting and conveying device of the pretreatment unit. When the kitchen waste material is sorted and conveyed by the conveyor belt, the near-infrared spectroscopy analyzer scans the material and estimates the moisture content and crude fat content of the current batch of material based on a pre-established near-infrared spectral model. When establishing the near-infrared spectral model, near-infrared reflectance spectral data under the same location and detection distance conditions are first acquired. The near-infrared reflectance spectral data includes the absorbance value sequence corresponding to multiple discrete wavelength points collected within a preset wavelength range and according to a preset spectral resolution, as well as the corresponding measurement environment temperature and the cumulative average number of scans. Then, the true moisture content and true crude fat content of the corresponding samples are determined by laboratory analysis methods as standard reference values. After pairing the near-infrared spectral data of the samples with the corresponding laboratory standard reference values, partial least squares regression is used to establish moisture content prediction models and crude fat content prediction models, respectively.

[0022] When receiving each batch of materials, operators enter the source category label of that batch through a human-machine interface, such as ordinary residential catering waste, group canteen waste, and aquatic seafood processing waste.

[0023] Real-time operating parameters include temperature, liquid level, pH value, etc., and this type of data can be directly acquired through the corresponding sensors.

[0024] Based on the crude fat content and source category label, combined with the statistical distribution characteristics of oil content in historical feed batches, the predicted oil load value after the current batch enters the anaerobic reactor is calculated.

[0025] After obtaining the feed parameters and real-time operating parameters, the oil loading characteristics of the material in the anaerobic reactor are predicted. This allows for the prediction of the current oil accumulation level inside the reactor before the anaerobic digestion reaction has fully progressed. Specifically, based on the source category label, the statistical distribution characteristics of oil content from the same category of source batches in past operating records are retrieved from the historical database. The statistical distribution characteristics of oil content are obtained by experimental measurement methods at multiple time points, and the statistical distribution characteristics are obtained by calculating the mean and standard deviation of the oil content. When the crude fat content measured by the near-infrared spectroscopy analyzer falls within the range of the mean plus or minus one standard deviation of the historical distribution of the same category, the crude fat content is directly used as the effective oil input. When the crude fat content deviates from the above range, the weighted average of the historical mean and the current crude fat content is used as the corrected effective oil input, with the current crude fat content given a lower weight to suppress possible deviations in near-infrared measurement under extremely uneven material conditions.

[0026] The effective oil input of the current batch is superimposed with the cumulative oil residue of the previous batches in the reactor. The result of this superposition is the predicted oil load in the anaerobic reactor at the current moment. The cumulative oil residue is calculated as follows: when the first batch of feed enters the reactor, the platform multiplies the effective oil input of that batch by the feed weight of that batch to obtain the absolute weight of oil carried into the reactor by the feed. This absolute weight of oil is recorded as the initial oil weight in the reactor. Subsequently, whenever a new batch of feed enters the reactor, the platform first needs to determine the remaining oil weight in the reactor after the previous batch was processed. The remaining oil weight is calculated as follows: based on the oil weight at the end of the previous batch, the weight of oil consumed and carried out in the reactor during the time interval from the previous batch to the current batch is deducted. The consumed and carried-out oil weight includes the amount of oil biodegraded by anaerobic microorganisms and the oil carried out by the anaerobic reactor effluent.

[0027] The method for calculating the biodegradation of lipids by anaerobic microorganisms involves establishing an empirical table of graded lipid degradation rates, using reactor temperature, pH, and lipid encapsulation level as input parameters. During system commissioning, batch anaerobic degradation experiments were conducted under different lipid encapsulation levels to measure the actual rate of decrease in lipid concentration over time at each encapsulation level, thus establishing an empirical table of degradation rates including corrections for lipid encapsulation level. In actual operation, before the system completes its initial lipid encapsulation level assessment, the lowest degradation rate corresponding to the highest encapsulation level is used for initial calculations to ensure that the residual lipid content is not underestimated. After the system completes its initial and subsequent lipid encapsulation level assessments, based on the lipid encapsulation level obtained from the previous batch assessment, combined with the reactor's average temperature and average pH during that time period, the corresponding corrected degradation rate is retrieved from the graded lipid degradation rate empirical table. This corrected degradation rate is multiplied by the time interval from the previous batch feed to the current batch feed to obtain the lipid concentration degradation amount during that time period. This lipid concentration degradation amount is then multiplied by the average volume of the anaerobic reactor during that time period, and the result is converted to units to obtain the biodegradation amount in mass during that time period.

[0028] The amount of oil carried out by the anaerobic reactor effluent is obtained by multiplying the effluent flow rate and the oil concentration in the effluent over a time interval. The effluent flow rate is calculated from the cumulative flow rate of the feed pump and changes in the liquid level. The effluent oil concentration is the average oil concentration obtained from periodic sampling and testing of the anaerobic effluent during historical operation. The sum of the biodegradation consumption and the amount carried out by the effluent is taken as the total oil weight loss for that time period. The oil weight remaining at the end of the previous batch of feed is obtained. Subtracting the total weight loss from the oil weight remaining yields the remaining oil weight in the reactor before the current new batch of feed. Subsequently, the effective oil input of the current new batch is multiplied by the feed weight of the previous batch to obtain the absolute weight of oil carried in by the current batch. This absolute weight of oil is added to the aforementioned remaining oil weight to obtain the total oil weight in the reactor at the current moment. Finally, the current effective volume of the reactor, calculated from the current liquid level, is read. Dividing the total oil weight by this current effective volume yields the predicted oil load in the anaerobic reactor at the current moment.

[0029] The oil load characteristics are obtained by correlating and correcting the predicted oil load values ​​based on temperature.

[0030] Since temperature changes affect the physical properties of oils (such as viscosity and fluidity) rather than their absolute mass, temperature-related corrections do not directly change the numerical value of the oil load forecast itself. Instead, they characterize the tendency of oils to coat solid substrates at different temperatures by introducing a temperature-coating risk coefficient.

[0031] The temperature-induced encapsulation risk coefficient is determined based on the deviation between the collected real-time temperature and the target temperature range for mesophilic fermentation. Specifically, when the real-time temperature is within the target temperature range of 35℃ to 38℃, the temperature-induced encapsulation risk coefficient is set to 1. When the real-time temperature is below 35℃ but not below 32℃, the temperature-induced encapsulation risk coefficient is set to 1.2, indicating that the encapsulation tendency of the oil at this temperature is increased by 20% compared to the target temperature range. When the real-time temperature is below 32℃, the temperature-induced encapsulation risk coefficient is set to 1.2 plus an increment of 0.05 for every degree Celsius below 32℃, but the upper limit of the temperature-induced encapsulation risk coefficient does not exceed 1.5 to avoid over-correction leading to an overreaction in subsequent control strategies. When the real-time temperature is above 38℃, the temperature-induced encapsulation risk coefficient is set to 0.9, indicating that the oil's fluidity is enhanced and its encapsulation tendency is reduced at higher temperatures. The oil load characteristics are composed of the predicted oil load value and the temperature-induced encapsulation risk coefficient.

[0032] In this embodiment, the calculation of the grease encapsulation level of the solid substrate based on the response sequence includes: The variable speed shearing process of the stirring device is divided into a low-shear zone, a medium-shear zone, and a high-shear zone. The execution parameters of the variable speed shearing are determined according to the characteristics of the oil load. The execution parameters include the scanning interval and the gear density of the speed gradient sequence of each shearing zone.

[0033] The scan trigger interval controls the waiting time between two adjacent variable-speed shear scan operations and is the execution frequency of the variable-speed shear scan operation. Speed ​​density refers to the number of speed speeds set within a specified shear zone in the speed gradient sequence. High-load and low-load thresholds are preset. The specific values ​​of the high-load and low-load thresholds are determined based on the anaerobic reactor's volume, design load, and empirical data from previous commissioning and operation.

[0034] The equivalent oil load value is obtained by multiplying the predicted oil load value by the temperature encapsulation risk coefficient. When the scanning interval is reached, if the equivalent oil load value exceeds the high load threshold, it indicates that the oil accumulation level in the reactor is high, and the risk of oil encapsulating the solid substrate is greater, requiring more frequent and precise detection of changes in the rheological properties of the material. At this time, the subsequent scanning interval is shortened to half of the baseline scanning interval, and the gear density of the rotational speed gradient sequence in the middle shear zone is increased, thereby increasing the detection frequency of the system and the accuracy of each detection. The baseline scanning interval is the standard scanning interval used by the system when the equivalent oil load value is between the high load threshold and the low load threshold. In this embodiment, the baseline scanning interval is set to 4 hours. Increasing the gear density in the middle shear zone means inserting more rotational speed gears in the middle shear zone of the rotational speed gradient sequence. For example, under the standard gear density, there are 4 rotational speed gears in the middle shear zone, and after increasing, there are 7 rotational speed gears. The reason for increasing the gear density in the middle shear zone is that the difference in the degree of oil encapsulation is mainly reflected in the torque response characteristics of the middle shear zone. A denser gear setting can more precisely capture the small shifts in the position of the shear thinning inflection point.

[0035] When the equivalent oil load value is lower than the low load threshold, it indicates a low level of oil accumulation in the reactor and a low risk of oil encapsulating the solid substrate. In this case, the scanning interval is extended to 1.5 times the baseline scanning interval to reduce the impact on conventional stirring, and a standard speed density is used. Under the standard speed density, the rotational speed gradient sequence is set with 3 speed levels in the low shear zone, 4 speed levels in the medium shear zone, and 3 speed levels in the high shear zone.

[0036] When the oil load characteristics are between the high load threshold and the low load threshold, variable speed shearing is performed using the reference scanning interval and standard gear density.

[0037] During the stirring intervals of the anaerobic reactor, the stirring device is controlled to operate according to the scan interval and speed gradient sequence based on the determined execution parameters.

[0038] After determining the execution parameters, wait for the anaerobic reactor to enter the regular stirring interval. The regular stirring interval refers to the rest period after the anaerobic reactor completes one round of regular stirring according to the preset stirring program during normal operation, and before entering the next round of regular stirring. Choosing to perform variable-speed shearing during the regular stirring interval is to avoid interference from the residual flow field of regular stirring on the torque measurement values, ensuring that the collected torque data reflects the true rheological response of the material when it is re-sheared under relatively static conditions.

[0039] After the conventional stirring stops, a preset settling time is allowed. In this embodiment, the settling time is set to 3 minutes to allow the macroscopic flow of materials in the reactor to essentially dissipate. After the settling time is completed, the determined scanning interval is used as the execution cycle for variable-speed shearing. The time when the last variable-speed shearing was completed is recorded. When the time elapsed since the last completion time reaches the scanning interval, the next variable-speed shearing is triggered. When the oil load characteristic value is updated due to a new batch of feed, the scanning interval is redefined, and subsequent variable-speed shearing is executed according to the redefined scanning interval. When the system is first started and no variable-speed shearing has been completed, the first variable-speed shearing is triggered immediately, regardless of the scanning interval.

[0040] The stirring device is started at the first speed setting in the speed gradient sequence. In this embodiment, the speed gradient sequence is arranged in ascending order, with the first speed setting being the lowest speed in the low-shear zone, for example, set to 20 revolutions per minute. After the stirring device operates at the first speed setting for a preset holding time, it switches to the second speed setting in the speed gradient sequence to continue operating. The holding time for this single setting is set to 90 seconds to ensure that the stirring device reaches a steady-state operation at each speed setting. This process is repeated sequentially, increasing the speed setting until the last speed setting in the speed gradient sequence has been completed.

[0041] When the stirring device is running at each speed level, the torque sensor collects the corresponding torque value, and all speed levels and their corresponding torque values ​​are arranged in sequence to form a response sequence.

[0042] During the operation of the stirring device according to the speed gradient sequence, a torque sensor installed on the stirring shaft collects the torque value borne by the stirring shaft. In this embodiment, a strain gauge torque sensor is used, installed at the coupling between the stirring shaft and the output end of the reducer, and the sampling frequency is set to 10 Hz. During the single-speed holding time of each speed gear, the torque value collected by the torque sensor will fluctuate in the initial stage due to the acceleration of the stirring device and the re-establishment of the material flow field. When the stirring device reaches a steady state at the current speed gear, the torque value tends to stabilize. To obtain the steady-state torque value, the torque collection data of the second third of the single-speed holding time of each speed gear is taken, and its arithmetic mean is calculated. The obtained arithmetic mean is used as the torque value corresponding to the current speed gear. The data of the second third of the time period is used to exclude the unstable data of the acceleration transition phase when the stirring device switches from the previous speed gear to the current speed gear.

[0043] After all speed gears have been operated, the speed value of each speed gear is used as the horizontal axis coordinate, and the corresponding steady-state torque value of each speed gear is used as the vertical axis coordinate. They are arranged in order from low to high speed to form a response sequence of torque as speed changes.

[0044] It should be noted that the torque value collected by the stirring shaft torque sensor reflects the overall macroscopic rheological properties of the mixture within the reactor, rather than the microscopic mechanical characteristics of the grease film on the surface of individual particles. When significant grease encapsulation occurs within the reactor, a large number of grease-encapsulated solid particles will form larger grease-solid composite flocs through the binding effect of the grease. The formation of these composite flocs alters the particle size distribution and effective solid volume fraction of the mixture, leading to detectable systematic changes in the overall apparent viscosity, yield stress, and shear thinning properties of the mixture. Therefore, this method detects the systematic shift in the macroscopic rheological properties of the mixture caused by grease encapsulation, rather than directly measuring the thickness of the grease film on the surface of individual particles.

[0045] Feature parameters are extracted from the response sequence. These feature parameters include the slope of the low-shear region, the inflection point of the medium-shear region, and the maximum difference of the high-shear region.

[0046] The slope of the low-shear zone is obtained by linearly fitting all data points in the low-shear zone using the least squares method; the slope of the fitted line is the desired slope. This slope reflects how quickly the torque of the material changes with increasing rotational speed at low shear rates. When the degree of grease coating in the reactor is high, the solid particles coated with grease in the material exhibit higher apparent viscosity under low-shear conditions, resulting in a faster rate of torque increase with rotational speed and a larger slope. When the degree of grease coating in the reactor is low, the flow resistance of the material is smaller, and the slope is relatively smaller.

[0047] The inflection point is the first point in the entire response sequence where the flow transitions from an upward convex to a downward concave shape. This inflection point reflects the critical shear rate at which the material transforms from approximately Newtonian fluid behavior to shear-thinning behavior. When the degree of grease coating within the reactor is high, the grease coating begins to peel and disperse at lower shear rates, and the material exhibits shear-thinning characteristics earlier, with the inflection point shifting towards lower rotational speeds. When the degree of grease coating is low, the inflection point is located in a higher rotational speed range. The calculation method for the inflection point is a conventional mathematical method and will not be described further here.

[0048] The maximum difference is calculated by subtracting the torque value corresponding to the lowest speed setting in the high-shear zone from the torque value corresponding to the highest speed setting in the high-shear zone. This maximum difference reflects the extent to which the torque increases when the shear rate continues to increase after the material has undergone sufficient shear thinning. When the grease coating in the reactor is high, some tightly coated solid particles are not completely stripped even in the medium-shear zone, generating additional flow resistance when the speed is further increased in the high-shear zone, resulting in a larger residual torque increment. When the grease coating is low, the material has essentially completed the shear thinning process in the high-shear zone, the torque increases more gradually with increasing speed, and the residual torque increment is smaller.

[0049] The characteristic parameters are compared with pre-calibrated reference characteristics to determine the grease coating level of the material in the current reactor.

[0050] After extracting three characteristic parameters, these parameters are compared with pre-calibrated reference characteristics to determine the grease coating level of the material in the reactor. The reference characteristics are obtained during the calibration phase before formal operation, by preparing simulated food waste and grease mixtures with different known weight ratios. These known weight ratios cover a continuous gradient from no grease coating to heavy grease coating. Each mixture sample is placed under the same temperature conditions as the actual operation, and the same rotational speed gradient sequence operation as described above is performed to collect the corresponding standard response sequence for each mixture sample. For each standard response sequence, three characteristic parameters are extracted using the same method as in step four, and their corresponding known grease coating levels are labeled. For example, if the grease coating level is divided into four levels, a grease to solid substrate weight ratio of 0% to 5% corresponds to level 1, 5% to 15% to level 2, 15% to 25% to level 3, and above 25% to level 4. All standard response sequences and their corresponding characteristic parameter values ​​and grease coating level labels are summarized to form the reference characteristics.

[0051] During the actual comparison, the three feature parameters extracted from the current response sequence are used to form a feature vector, and the Euclidean distance between this vector and the reference feature is calculated. Specifically, the three features need to be normalized before calculating the Euclidean distance. The grease coating level corresponding to the reference feature with the smallest Euclidean distance is taken as the grease coating level of the material in the current reactor.

[0052] In this embodiment, the functional status diagnosis results of acid-producing bacteria are obtained based on the trajectory morphology features of multidimensional vector sequences in the component space, including: At each sampling time, the concentration values ​​of various volatile fatty acids in the effluent of the anaerobic reactor were collected. After normalization of the concentration values, a multi-component concentration data with a sum of 1 was constructed.

[0053] The volatile fatty acids in this embodiment include acetic acid, propionic acid, butyric acid, and valeric acid. Sampling was performed every 4 hours. First, the concentration values ​​of the four volatile fatty acids were processed using L1 normalization to ensure that the sum of their normalized proportions was 1. Since the four components after normalization are subject to a closure constraint of a sum of 1, there are negative correlations between the components due to mathematical constraints rather than biochemical mechanisms. Directly calculating the Euclidean distance and geometric centroid in the closed space would introduce spurious statistical associations. Therefore, a central logarithmic ratio transformation was performed on the normalized concentration data to eliminate the closure effect before constructing the multi-component concentration data.

[0054] The specific method of central logarithmic ratio transformation is as follows: For the normalized four-component concentration data, where the four components represent the normalized proportions of acetic acid, propionic acid, butyric acid, and valeric acid, respectively, the geometric mean of the four components is first calculated, which is the fourth root of the product of the four components. Then, the natural logarithm of the ratio of each component to the geometric mean is calculated, resulting in the transformed four-dimensional vector. When the normalized proportion of a component is zero, this zero-value component is replaced with 0.001 before calculation, and the four components after replacement are re-L1 normalized to restore the sum to 1 before performing the central logarithmic ratio transformation. The four-dimensional vector after the central logarithmic ratio transformation is no longer subject to the closure constraint that the sum equals 1, and the correlation between the components reflects the actual biochemical metabolic associations. The four-dimensional vector after the central logarithmic ratio transformation is used as the multi-component concentration data at that sampling time.

[0055] Set a sliding time window and connect the multi-component concentration data within the sliding time window in chronological order to obtain a multidimensional vector sequence.

[0056] In this embodiment, the sliding time window is set to 48 hours. This 48-hour window is chosen because functional shifts in acid-producing bacteria typically require one to two complete hydraulic retention time cycles to produce identifiable trend changes in the volatile fatty acid composition structure in the early stages. Generally, the hydraulic retention time in an anaerobic reactor is set to 20 to 25 days. The 48-hour window span accounts for approximately 8% to 10% of the hydraulic retention time, which, at this scale, smooths out random fluctuations from a single sampling while preserving trend information on changes in bacterial community function. The multi-component concentration data collected at the most recent moment within the sliding time window are concatenated chronologically to obtain a multidimensional vector sequence.

[0057] Each multi-component concentration data in the multidimensional vector sequence is mapped to a corresponding coordinate point in the component space, and the coordinate points are connected in chronological order to obtain the component trajectory.

[0058] The component space is a four-dimensional Euclidean space with the central logarithmic ratio transformation components of acetic acid, propionic acid, butyric acid, and valeric acid as the four coordinate axes. Based on the acetic acid, propionic acid, butyric acid, and valeric acid component values ​​of the multi-component concentration data after central logarithmic ratio transformation, the multi-component concentration data are mapped to the corresponding coordinate points in this space.

[0059] Trajectory morphology features are extracted from the component trajectories. These features include the centroid drift direction and drift rate, the trend of the spread radius, and the projection residence density distribution of the component trajectories.

[0060] The direction of centroid drift reflects the shift in the acid-producing pathway. For example, a rapid drift toward acetic acid and propionic acid often indicates that the microbial community is overly shifting toward the metabolism of easily degradable sugars, while the drift rate measures the urgency of this functional aberration.

[0061] The trend of the distribution radius reflects the stability and concentration of the system's metabolic state. When the distribution radius shows a shrinking trend, it means that the proportion of volatile fatty acids is gradually locked in a certain pattern, and the fluctuation range is reduced. This is usually a characteristic manifestation of the microbial community's metabolic pathways becoming more singular and its ability to degrade complex substrates declining. Conversely, the expansion or stabilization of the radius suggests the diversity of metabolic pathways.

[0062] The distribution of projected residence density, particularly its characteristics on the propionic-butyric acid subplane, is crucial. Due to the high oil content in food waste, the oil-coated substrate significantly inhibits the β-oxidation of fatty acids, with this inhibition often manifesting most prominently in the abnormal accumulation of propionic acid. Calculating the projected residence density allows identification of whether high-density regions are concentrated in specific risk zones of high propionic acid and low butyric acid. If the trajectory remains in this region for an extended period and its range shrinks, it constitutes a characteristic signal of impaired fatty acid metabolism.

[0063] The trajectory morphology features are input into the pre-classification model to obtain the state diagnosis results.

[0064] The trajectory morphology features are combined with the propionic acid-butyric acid drift ratio to form a feature input vector. When combining the feature input vector, the four components of the centroid drift direction, the drift rate, and the change in the spread radius are arranged sequentially. Then, the residence density matrix is ​​flattened into a one-dimensional vector in row-major order and concatenated after it. Finally, the propionic acid-butyric acid drift ratio is concatenated to form a complete feature input vector. The propionic acid-butyric acid drift ratio is calculated as follows: the absolute values ​​of the propionic acid component and the butyric acid component in the centroid drift direction are extracted. When the absolute value of the butyric acid component is greater than 0.001, the propionic acid-butyric acid drift ratio is equal to the absolute value of the propionic acid component divided by the absolute value of the butyric acid component. When the absolute value of the butyric acid component is not greater than 0.001, the propionic acid-butyric acid drift ratio is set to 10. The propionate-butyrate drift ratio is used to distinguish between two different drift patterns: carbohydrate metabolism shift and fatty acid metabolism inhibition. When the rate of change of propionate percentage is significantly faster than the rate of change of butyrate percentage, the propionate-butyrate drift ratio takes a larger value, indicating that the fatty acid β-oxidation pathway is blocked, leading to abnormal accumulation of propionate.

[0065] Before input, the continuous components in the feature input vector are subjected to max-min normalization. The normalized feature input vector is then input into a pre-trained classification model. The training data for the classification model is obtained by applying different feed group ratios and stirring conditions to the anaerobic reactor during the debugging phase, subjecting the reactor to four operating conditions: functional equilibrium, carbohydrate metabolism deviation, fatty acid metabolism inhibition, and compound deviation warning. Under each condition, multi-component concentration data of volatile fatty acids are collected, and the feature input vector is calculated using the aforementioned method. Simultaneously, the functional state of the acid-producing bacteria is confirmed through laboratory 16S rRNA gene sequencing analysis, and the confirmed functional state is used as the label for the corresponding feature input vector. In this embodiment, the classification model uses a random forest algorithm, with 200 decision trees, a maximum tree depth of 8 layers, and a minimum number of leaf node samples of 5.

[0066] The classification model outputs the current state diagnosis results of the acid-producing bacteria community. The state diagnosis results include functional equilibrium state, carbohydrate metabolism deviation state, fatty acid metabolism inhibition state, and compound deviation warning state. Among them, the functional equilibrium state means that the gas production efficiency and organic matter removal rate of the anaerobic reactor are at the design level, there is no accumulation of a large amount of unacidified substrate in the anaerobic effluent, and the influent load of the downstream aerobic treatment unit is within the normal range.

[0067] In a shifted state of carbohydrate metabolism, the proportions of acetic acid and propionic acid will increase simultaneously, while the proportion of butyric acid will decrease. If not intervened in time, fatty acids and protein substrates in the reactor will gradually accumulate and eventually enter the downstream aerobic treatment unit in an unacidified form with the effluent, causing the organic load of the aerobic system to exceed the standard.

[0068] Under the state of inhibited fatty acid metabolism, long-chain fatty acids accumulate. These long-chain fatty acids have a toxic inhibitory effect on methanogenic archaea, leading to a decrease in gas production and methane content. At the same time, a large amount of undegraded oils and long-chain fatty acids enter the aerobic system with the effluent, causing foaming in the aeration tank, a decrease in dissolved oxygen mass transfer efficiency, and the formation of hydrophobic flocs in the activated sludge.

[0069] In the state of compound shift warning, although the various characteristic indicators have not yet reached the complete judgment criteria for a single shift state, a shift trend has already appeared. If not intervened in time, the microbial community status will deteriorate into one of the following sampling window periods: a carbohydrate metabolism shift state or a fatty acid metabolism inhibition state, or both, resulting in a severe imbalance.

[0070] In this embodiment, the trajectory morphology features extracted from the component trajectories include: The sliding time window is divided into a first half sub-window and a second half sub-window. The geometric centroids of all multi-component concentration data in the first half sub-window and the second half sub-window are calculated respectively, and used as the early centroid and the late centroid respectively.

[0071] Assuming the first half of the sub-window includes six multi-component concentration data points, and since each multi-component concentration data point comprises four data points, it is vector data. The arithmetic mean of each of the four components of the six multi-component concentration data points is calculated. The combination of these four averages—the arithmetic mean of the logarithmic ratio transformation components of acetic acid, propionic acid, butyric acid, and valeric acid centers—is the early centroid. Similarly, the later centroid is calculated using this method.

[0072] The direction vector pointing from the early centroid to the later centroid is used as the centroid drift direction, and the drift rate is calculated based on the magnitude of the direction vector and the time interval between the first and second half sub-windows.

[0073] The centroid drift direction is obtained by subtracting the coordinates of the earlier centroid from the coordinates of the later centroid. The magnitude of the centroid drift direction is calculated, and the drift rate is obtained by dividing the magnitude by the time interval between the first and second half of the sub-window. The time interval between the first and second half of the sub-window is half the duration of the sliding time window; in this example, it is 24 hours. The centroid calculated in the above steps represents the average state of the proportion of each component within the corresponding time window, while the centroid drift direction reflects the shift target of the acid production pathway. For example, a drift towards acetic acid and propionic acid indicates that the microbial community is transitioning excessively towards the metabolism of easily degradable sugars, and the drift rate quantifies the urgency of this shift.

[0074] Calculate the average Euclidean distance from each multi-component concentration data point in the first half of the sub-window to the centroid of the previous period to obtain the initial spread radius. Calculate the average Euclidean distance from each multi-component concentration data point in the second half of the sub-window to the centroid of the later period to obtain the later spread radius. Use the difference between the later spread radius and the initial spread radius as the spread radius change value.

[0075] Define the centroid within the first half of the sub-window as the early centroid and the centroid within the second half of the sub-window as the late centroid. Calculate the Euclidean distance from each multi-component concentration data point within the first half of the sub-window to the early centroid, and then calculate the average of all Euclidean distances within the first half of the sub-window to obtain the early distribution radius. Similarly, calculate the Euclidean distance from each multi-component concentration data point within the second half of the sub-window to the late centroid, and then calculate the average of all Euclidean distances within the second half of the sub-window to obtain the late distribution radius. Subtract the early distribution radius from the late distribution radius to obtain the change in distribution radius.

[0076] When the change in the spread radius is negative, it indicates that the spread range of the component trajectory in the second half is shrinking compared to the first half. In other words, the multi-component concentration data at each sampling time in the second half are more concentrated relative to their centroid, indicating that the metabolic activity of the acid-producing bacteria is locked onto a certain type of substrate.

[0077] When the change in the spread radius is positive, it indicates that the spread range is expanding. That is, the multi-component concentration data at each sampling time in the latter half are more dispersed relative to their centroid, indicating that multiple substrates enter the degradation process at the same time and the metabolic activities of acid-producing bacteria alternate between multiple substrates.

[0078] When the change in the spread radius is close to zero, it indicates that there is no significant difference in metabolic activity between the preceding and following time periods.

[0079] The multi-component concentration data after central logarithmic ratio transformation are projected onto a two-dimensional subplane with the central logarithmic ratio transformation component of propionic acid as the horizontal axis and the central logarithmic ratio transformation component of butyric acid as the vertical axis to obtain a sequence of projection points. The two-dimensional subplane is divided into multiple grid cells, and the number of times the projection point sequence resides in each grid cell is counted to obtain a residence density map.

[0080] The direction of centroid drift, drift rate, change in spread radius, and dwell density map are used as trajectory morphology features.

[0081] The reason for choosing propionic acid and butyric acid as projection planes is that, in anaerobic digestion, the accumulation of propionic acid reflects the inhibition of the fatty acid β-oxidation pathway, while the change in butyric acid concentration is related to the transformation of intermediate metabolites in fatty acid degradation. The change in the ratio of propionic acid to butyric acid has good sensitivity in distinguishing whether fatty acid metabolism is inhibited.

[0082] In the specific calculation, the multi-component concentration data of each component trajectory after central logarithmic ratio transformation is projected onto the propionic acid-butyric acid two-dimensional subplane. During projection, for the multi-component concentration data within the sliding time window, the propionic acid component and butyric acid component after central logarithmic ratio transformation are extracted to form a projection point. In this way, all multi-component concentration data within the sliding time window are projected to obtain the projection point sequence.

[0083] The two-dimensional subplane is then divided into a non-uniform grid. Since the values ​​of the propionic acid and butyric acid components after the central logarithmic ratio transformation are no longer limited to 0 to 1, but can take any real value, it is necessary to determine the coordinate range of the projection plane based on historical data collected during the system debugging phase. Specifically, the minimum and maximum values ​​of the propionic acid and butyric acid components after the central logarithmic ratio transformation are statistically analyzed at all sampling times during the system debugging phase, and then extended outwards by 10% to define the effective ranges for the propionic acid and butyric acid axes, respectively. In this embodiment, based on the debugging phase data, the effective range of the propionic acid axis is determined to be -2.5 to 2.5, and the effective range of the butyric acid axis is -2 to 2.

[0084] like Figure 2 As shown, a non-uniform grid partitioning strategy is adopted to balance the fine resolution of the normal operating range and the coverage of the abnormal range. On the propionic acid axis, the effective range is divided into three segments: the segment from -2.5 to -0.5 is divided into 4 equal-width grid cells, the segment from -0.5 to 1 is divided into 9 equal-width grid cells, and the segment from 1 to 2.5 is divided into 3 equal-width grid cells. On the butyric acid axis, the effective range is divided into three segments: the segment from -2 to -0.5 is divided into 3 equal-width grid cells, the segment from -0.5 to 0.8 is divided into 8 equal-width grid cells, and the segment from 0.8 to 2 is divided into 2 equal-width grid cells. This partitioning ensures that the middle segment, where normal operating data is usually concentrated, has the highest grid density, enabling fine differentiation of subtle differences in normal fluctuation patterns. Although the grid in the abnormal range is coarser, it still records the existence of projection points and does not lose characteristic information of extreme abnormal states.

[0085] Iterate through each projection point in the projection point sequence, determining its row and column number within the grid cell based on its propionic and butyric coordinates. When a projection point's propionic or butyric coordinates exceed the valid range, assign it to the outermost grid cell of that axis. Count the number of projection points included in each grid cell to obtain the residency count matrix. Divide each element in the residency count matrix by the total number of projection points to obtain the residency density matrix. Each element in the residency density matrix ranges from 0 to 1, representing the relative frequency of the corresponding grid cell being occupied by projection points.

[0086] In this embodiment, the generation of stirring strategy adjustment instructions and feed control instructions includes: The particle swarm optimization algorithm generates adjustment instructions for the stirring strategy and control instructions for the feeding. When the particle swarm optimization algorithm is executed, a control parameter space is defined, and the particle swarm is initialized in the control parameter space. The position vector of each particle represents the combination of parameter control.

[0087] In this embodiment, the control parameter space includes stirring speed ratio, pulse duration, adjustment amount of carbohydrate substrate ratio, adjustment amount of lipid content ratio, and adjustment ratio of total feed load. Specifically, the stirring speed ratio ranges from 1 to 3, the pulse duration ranges from 0 to 300 seconds, the carbohydrate substrate ratio adjustment amount ranges from -0.3 to 0.3, the lipid content ratio adjustment amount ranges from -0.3 to 0.3, and the total feed load adjustment ratio ranges from -0.3 to 0.2.

[0088] The particle swarm is initialized in the parameter space, with a size of 30 particles. The position vector of each particle is a five-dimensional vector, representing a set of parameter control combinations. The initial position of each particle is randomly and uniformly generated within the range of values ​​in each dimension. The initial velocity of each particle is randomly and uniformly generated within the velocity limit range in each dimension, with the velocity limit value in each dimension set to 20% of the width of the range of values ​​in that dimension.

[0089] A fitness function is constructed based on the current oil encapsulation level and the current functional state category of the acid-producing bacteria. The fitness value is calculated based on the fitness function, and the velocity and position of each particle are updated.

[0090] The fitness function is used to evaluate the regulatory effect of the parameter combination represented by each particle under the current oil encapsulation level and functional state category. The fitness function consists of two parts: an improvement benefit term and a coupling risk penalty term. The specific formula of the fitness function is as follows: Where 5 is the penalty weighting coefficient. For fitness values, To improve efficiency, As a coupling risk penalty term, a larger fitness value indicates a better combination of parameters.

[0091] When calculating the improvement benefits, the system first consults a pre-defined agitation-oil dispersion efficiency mapping table based on the stirring speed ratio and pulse duration at the particle location. The estimated oil dispersion efficiency is obtained from this table, which represents the dispersion capability of agitation parameters for standard materials. This mapping table is calibrated through controlled experiments during system debugging. The estimated oil dispersion efficiency is multiplied by the current oil coating level code value to obtain the agitation improvement score; for example, level 1 corresponds to code value 1, and level 2 corresponds to code value 2. The agitation improvement score reflects the greater the improvement in condition resulting from effective agitation and dispersion under conditions of more severe oil coating.

[0092] Adjust the amount based on the proportion of carbohydrate substrate at the particle location. Adjustment amount of fat-containing materials And the current functional status category calculates the feed improvement score. When the functional state category is carbohydrate metabolism shifted, a decrease in the proportion of carbohydrate substrates helps suppress the competitive advantage of carbohydrate-metabolizing microbiota, thus improving the feed improvement score. , When it is negative A positive value indicates a greater reduction in the feed rate, resulting in better improvement. When the functional state is in a state of functional equilibrium, maintaining the existing feed ratio is the optimal strategy, and the feed improvement score is [value missing]. , Adjusting the total feed load ratio in any direction will slightly reduce the improvement benefit. When the functional state category is fatty acid metabolism inhibition, reducing the proportion of lipid-containing materials helps alleviate the burden on the microbial community, and the feed improvement score will be lower in this case. , When it is negative The value is positive. When the functional status category is composite offset warning state, the feed improvement score is... The adjustment contributions from both carbohydrates and lipids are considered.

[0093] The improvement benefit item is the sum of the maximum and minimum normalized mixing improvement score and the feeding improvement score.

[0094] For the coupling risk penalty term, the estimated total fatty acid load corresponding to the particle position is first calculated. The estimated amount of fatty acid released by stirring is calculated by multiplying the estimated oil dispersion efficiency by the oil encapsulation level code value and the preset fatty acid release coefficient. The fatty acid release coefficient is calibrated experimentally during the system debugging phase.

[0095] Then, the estimated amount of fatty acids input into the feed is calculated. Calculated based on the following formula: ,in, This is the baseline amount of fatty acids, expressed in g / L. This parameter is calculated based on the predicted oil load of the current feed batch and the preset hydrolysis conversion coefficient from oil to fatty acids. The hydrolysis conversion coefficient was calibrated experimentally during the system commissioning phase. The total feed load is adjusted accordingly. Then, the estimated total fatty acid load is calculated, which is the sum of the estimated amount of fatty acids released during stirring and the estimated amount of fatty acids input into the feed.

[0096] Finally, obtain the fatty acid load limit corresponding to the current functional status category. The preset standard tolerance values ​​are as follows: In a functional equilibrium state, the upper limit of fatty acid load equals the preset standard tolerance value; in a carbohydrate metabolism deviation state, the upper limit of fatty acid load is 0.85 times the preset standard tolerance value; in a fatty acid metabolism inhibition state, the upper limit of fatty acid load is 0.7 times the preset standard tolerance value; and in a compound deviation warning state, the upper limit of fatty acid load is 0.6 times the preset standard tolerance value. When the estimated total fatty acid load is less than the upper limit of fatty acid load, the coupling risk penalty term is 0; otherwise, the coupling risk penalty term is... ,in, To estimate the total fatty acid load, This represents the upper limit of fatty acid load.

[0097] After the iteration terminates, the stirring-related parameters in the parameter control combination corresponding to the global optimal position are used to generate stirring strategy adjustment instructions, and the feeding-related parameters are used to generate feeding control instructions.

[0098] The standard particle swarm optimization algorithm is used for iterative updates of the particle swarm. In each iteration, the fitness value of each particle is calculated according to the fitness function. If the current fitness value of a particle is better than its historical best fitness value, the current position is updated to the historical best position of that particle. In this embodiment, the number of iterations is set to 100. The individual learning factor is set to 2, the global learning factor is set to 2, and the inertia weight adopts a linear decreasing strategy, decreasing linearly from an initial value of 0.9 to a final value of 0.4 with the number of iterations. After the position update is completed, the positions of particles that exceed the range of values ​​for each parameter are truncated. If the new position in a certain dimension exceeds the upper limit of the range of values ​​for that dimension, it is truncated to the upper limit value; if it is lower than the lower limit, it is truncated to the lower limit value.

[0099] The position with the highest fitness value among all the individual best positions of each particle is selected and updated as the global best position. The five-dimensional parameter combination corresponding to the global best position is then used as the optimal control parameter combination.

[0100] The stirring strategy adjustment command is generated based on the stirring speed multiplier and pulse duration in the optimal control parameter combination. The stirring strategy adjustment command includes adjusting the stirring device's speed setpoint to the normal speed multiplied by the stirring speed multiplier, and setting the single pulse stirring duration to the value of the pulse duration. When the pulse duration value is 0, the stirring strategy adjustment command maintains the normal continuous stirring mode.

[0101] Feed control instructions are generated based on the adjustments to the proportion of carbohydrate substrate, the proportion of lipids, and the total feed load in the optimal control parameter combination. These instructions include adjusting the proportion of carbohydrate substrate in the feed to the current proportion plus the carbohydrate substrate adjustment amount, adjusting the proportion of lipids to the current proportion plus the lipids adjustment amount, and adjusting the total feed load to the current load multiplied by the total feed load adjustment ratio plus 1.

[0102] like Figure 3 The diagram shown illustrates the effectiveness of this invention in a real-world application scenario. Figure 3 The invention demonstrates a comparison of key performance indicators between the method of this invention (experimental group) and the traditional macroscopic parameter feedback adjustment method (control group) in a food waste resource recovery project during a continuous operating cycle. Figure 3 The comparative results show that the present invention effectively solves the technical bottleneck of traditional methods that cannot detect latent aberrations in bacterial community function caused by the substrate being coated with grease in a timely manner through the synergistic effect of multidimensional data fusion diagnosis and intelligent control strategy, and significantly improves the operational stability and overall treatment efficiency of the kitchen waste sewage treatment system.

[0103] like Figure 4 As shown, the present invention also provides a food waste wastewater treatment system based on multidimensional data diagnosis. This system is used to implement the methods described above, and includes: The operating condition sensing module acquires the feed parameters of kitchen waste and the real-time operating parameters of the anaerobic reactor, and calculates the oil load characteristics of kitchen waste wastewater based on the feed parameters and real-time operating parameters.

[0104] The physical property detection module performs variable-speed shear scanning on the stirring device of the anaerobic reactor based on the characteristics of oil loading, and collects the response sequence of stirring torque as the rotation speed changes. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate is calculated.

[0105] The biochemical diagnostic module collects multi-component concentration data of volatile fatty acids in the effluent of the anaerobic reactor, constructs a multi-dimensional vector sequence from the multi-component concentration data within a continuous time window, and obtains the status diagnostic results of acid-producing bacteria based on the trajectory morphology characteristics of the multi-dimensional vector sequence in the component space.

[0106] The intelligent decision-making module generates adjustment instructions for the stirring strategy and control instructions for the feed based on the oil coating level and state diagnosis results. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

[0107] It should be noted that the specific numerical parameters, threshold settings, and algorithm hyperparameter configurations involved in the various embodiments of the present invention are all exemplary and do not constitute a limitation on the scope of protection of the present invention. Specifically, the various threshold parameters appearing in this specification, including but not limited to the high and low load thresholds for oil and fat load, the boundary values ​​for dividing oil and fat encapsulation levels, the segmented values ​​and upper limits of the temperature encapsulation risk coefficient, the corresponding multiples of each state category of the upper limit of fatty acid load, the cutoff threshold of 0.001 for the propionic acid-butyric acid drift ratio, and the zero-value substitution value of 0.001, can all be determined by those skilled in the art based on the volume specifications of the anaerobic reactor, the designed organic load, the characteristics of the feed composition, and the experimental data from the previous commissioning and operation in the specific application scenario, through conventional parameter calibration experiments and engineering experience. Different value selections do not affect the substantive implementation of the technical solution of the present invention. The algorithm parameters mentioned in this specification, including but not limited to particle swarm size, number of iterations, individual learning factor, global learning factor and velocity limit ratio in particle swarm optimization algorithm, number of decision trees, maximum tree depth and minimum number of leaf node samples in random forest classification model, and penalty weight coefficient in fitness function, are all conventional design choices that can be determined by those skilled in the art through cross-validation, grid search or other conventional hyperparameter tuning methods based on specific optimization objectives, computational resource constraints and classification accuracy requirements.

[0108] The normalization methods mentioned in this specification, including but not limited to L1 normalization of volatile fatty acid concentration values, normalization of feature parameters, maximum-minimum normalization of feature input vectors, and central logarithmic ratio transformation, are all standard technical means in the field of data analysis and pattern recognition. Those skilled in the art can choose equivalent normalization or data transformation methods to replace them according to the data distribution characteristics and subsequent analysis needs. For example, Z-score normalization can be used instead of maximum-minimum normalization, and additive logarithmic ratio transformation or equidistant logarithmic ratio transformation can be used instead of central logarithmic ratio transformation. Such alternative schemes do not exceed the scope of the technical concept of this invention.

[0109] Furthermore, other specific implementation details involved in this specification, including but not limited to sampling frequency, sampling interval, sliding time window length, single-gear holding time, static stabilization time, grid division method and segment boundary values, installation location and type of torque sensor, modeling method of near-infrared spectral model, etc., can all be adaptively adjusted and equivalently replaced by those skilled in the art according to actual engineering conditions and equipment configuration. As long as they do not deviate from the core concept of the technical solution defined by the claims of this invention, they should all be considered to fall within the protection scope of this invention.

[0110] In summary, the numerical values, parameters, and method selections given in the specific embodiments of this specification are intended to help those skilled in the art understand and implement the present invention, and should not be construed as limiting the only implementation of the technical solutions of the present invention. Based on the teachings of this specification, those skilled in the art can reasonably adjust and replace the above parameters and methods without creative effort, and the technical solutions obtained thereby still fall within the protection scope of the present invention.

Claims

1. A method for treating kitchen waste wastewater based on multidimensional data diagnosis, characterized in that, include: The feed parameters of food waste and the real-time operating parameters of the anaerobic reactor were obtained, and the oil load characteristics of the food waste wastewater were calculated based on the feed parameters and real-time operating parameters. Based on the characteristics of oil loading, a variable-speed shear scan was performed on the stirring device of the anaerobic reactor, and the response sequence of stirring torque as a function of rotational speed was collected. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate was calculated. Multi-component concentration data of volatile fatty acids in the effluent of an anaerobic reactor were collected. The multi-component concentration data within a continuous time window were constructed into a multi-dimensional vector sequence. Based on the trajectory morphology characteristics of the multi-dimensional vector sequence in the component space, the functional status diagnosis results of the acid-producing bacteria were obtained. Based on the oil coating level and state diagnosis results, adjustment instructions for the stirring strategy and control instructions for the feed are generated. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

2. The method according to claim 1, characterized in that, The oil load characteristics of kitchen waste wastewater are calculated based on feed parameters and real-time operating parameters, including: Feed parameters include weight, crude fat content, and source category label; real-time operating parameters include temperature and liquid level data. Based on the crude fat content and source category label, combined with the statistical distribution characteristics of oil content in historical feed batches, the predicted oil load value after the current batch enters the anaerobic reactor is calculated. The oil load characteristics are obtained by correlating and correcting the predicted oil load values ​​based on temperature.

3. The method according to claim 1, characterized in that, The degree of grease encapsulation on solid substrates is calculated based on response sequences, including: The variable speed shearing process of the stirring device is divided into a low shearing zone, a medium shearing zone, and a high shearing zone. The execution parameters of the variable speed shearing are determined according to the characteristics of the oil load. The execution parameters include the scanning interval and the gear density of the speed gradient sequence of each shearing zone. During the stirring interval of the anaerobic reactor, the stirring device is controlled to operate according to the scan interval and speed gradient sequence according to the determined execution parameters; When the stirring device is running at each speed level, the torque sensor collects the corresponding torque value, and all speed levels and corresponding torque values ​​are arranged in sequence to form a response sequence. Feature parameters are extracted from the response sequence, including the slope of the low shear region, the inflection point of the medium shear region, and the maximum difference of the high shear region. The characteristic parameters are compared with pre-calibrated reference characteristics to determine the grease coating level of the material in the current reactor.

4. The method according to claim 1, characterized in that, The functional status diagnosis results of acid-producing bacteria are obtained based on the trajectory morphology features of multidimensional vector sequences in the component space, including: At each sampling time, the concentration values ​​of various volatile fatty acids in the effluent of the anaerobic reactor were collected, and the concentration values ​​were normalized to construct a multi-component concentration data with a sum of 1. Set a sliding time window and connect the multi-component concentration data within the sliding time window in chronological order to obtain a multi-dimensional vector sequence; Each multi-component concentration data in the multidimensional vector sequence is mapped to a corresponding coordinate point in the component space, and the coordinate points are connected in chronological order to obtain the component trajectory; Trajectory morphology features are extracted from the component trajectories. These features include the centroid drift direction and drift rate, the trend of the spread radius, and the projection residence density distribution of the component trajectories. The trajectory morphology features are input into the pre-classification model to obtain the state diagnosis results.

5. The method according to claim 4, characterized in that, The state diagnosis results include functional equilibrium state, carbohydrate metabolism deviation state, fatty acid metabolism inhibition state, and compound deviation warning state.

6. The method according to claim 4, characterized in that, Extracting trajectory morphology features from component trajectories, including: The sliding time window is divided into a first half sub-window and a second half sub-window. The geometric centroids of all multi-component concentration data in the first half sub-window and the second half sub-window are calculated respectively, and used as the early centroid and the late centroid respectively. The direction vector pointing from the early centroid to the later centroid is used as the centroid drift direction, and the drift rate is calculated based on the magnitude of the direction vector and the time interval between the first half and the second half of the sub-window. Calculate the average Euclidean distance from each multi-component concentration data point in the first half of the sub-window to the centroid of the previous period to obtain the initial spread radius. Calculate the average Euclidean distance from each multi-component concentration data point in the second half of the sub-window to the centroid of the later period to obtain the later spread radius. Use the difference between the later spread radius and the initial spread radius as the spread radius change value. The multi-component concentration data after central log ratio transformation are projected onto a two-dimensional subplane with the central log ratio transformation component of propionic acid as the horizontal axis and the central log ratio transformation component of butyric acid as the vertical axis to obtain the projection point sequence. The two-dimensional subplane is divided into multiple grid cells, and the number of times the projection point sequence resides in each grid cell is counted to obtain the residence density map. The direction of centroid drift, drift rate, change in spread radius, and dwell density map are used as trajectory morphology features.

7. The method according to claim 1, characterized in that, Generate adjustment instructions for the mixing strategy and control instructions for the feed, including: The particle swarm optimization algorithm generates adjustment instructions for the stirring strategy and control instructions for the feeding. When the particle swarm optimization algorithm is executed, a control parameter space is defined, and the particle swarm is initialized in the control parameter space. The position vector of each particle represents the combination of parameter control. A fitness function is constructed based on the current oil encapsulation level and the current functional state category of the acid-producing bacteria. The fitness value is calculated based on the fitness function, and the velocity and position of each particle are updated. After the iteration terminates, the stirring-related parameters in the parameter control combination corresponding to the global optimal position are used to generate stirring strategy adjustment instructions, and the feeding-related parameters are used to generate feeding control instructions.

8. The method according to claim 7, characterized in that, The control parameter space includes stirring speed ratio, pulse duration, adjustment amount of sugar substrate ratio, adjustment amount of lipid material ratio, and adjustment ratio of total feed load.

9. A kitchen waste and wastewater treatment system based on multidimensional data diagnosis, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: The operating condition sensing module acquires the feed parameters of kitchen waste and the real-time operating parameters of the anaerobic reactor, and calculates the oil load characteristics of kitchen waste wastewater based on the feed parameters and real-time operating parameters. The physical property detection module performs variable-speed shear scanning on the stirring device of the anaerobic reactor based on the characteristics of oil loading, and collects the response sequence of stirring torque as the speed changes. Based on the response sequence, the oil encapsulation level of the oil on the solid substrate is calculated. The biochemical diagnostic module collects multi-component concentration data of volatile fatty acids in the effluent of the anaerobic reactor, constructs multi-dimensional vector sequences from the multi-component concentration data within a continuous time window, and obtains the functional status diagnostic results of acid-producing bacteria based on the trajectory morphology characteristics of the multi-dimensional vector sequences in the component space. The intelligent decision-making module generates adjustment instructions for the stirring strategy and control instructions for the feed based on the oil coating level and state diagnosis results. The adjustment instructions and control instructions are used to regulate the stirring parameters and feed group distribution ratio of the anaerobic reactor, respectively.

Images